DonorsChoose

DonorsChoose.org receives hundreds of thousands of project proposals each year for classroom projects in need of funding. Right now, a large number of volunteers is needed to manually screen each submission before it's approved to be posted on the DonorsChoose.org website.

Next year, DonorsChoose.org expects to receive close to 500,000 project proposals. As a result, there are three main problems they need to solve:

  • How to scale current manual processes and resources to screen 500,000 projects so that they can be posted as quickly and as efficiently as possible
  • How to increase the consistency of project vetting across different volunteers to improve the experience for teachers
  • How to focus volunteer time on the applications that need the most assistance

The goal of the competition is to predict whether or not a DonorsChoose.org project proposal submitted by a teacher will be approved, using the text of project descriptions as well as additional metadata about the project, teacher, and school. DonorsChoose.org can then use this information to identify projects most likely to need further review before approval.

About the DonorsChoose Data Set

The train.csv data set provided by DonorsChoose contains the following features:

Feature Description
project_id A unique identifier for the proposed project. Example: p036502
project_title Title of the project. Examples:
  • Art Will Make You Happy!
  • First Grade Fun
project_grade_category Grade level of students for which the project is targeted. One of the following enumerated values:
  • Grades PreK-2
  • Grades 3-5
  • Grades 6-8
  • Grades 9-12
project_subject_categories One or more (comma-separated) subject categories for the project from the following enumerated list of values:
  • Applied Learning
  • Care & Hunger
  • Health & Sports
  • History & Civics
  • Literacy & Language
  • Math & Science
  • Music & The Arts
  • Special Needs
  • Warmth

Examples:
  • Music & The Arts
  • Literacy & Language, Math & Science
school_state State where school is located (Two-letter U.S. postal code). Example: WY
project_subject_subcategories One or more (comma-separated) subject subcategories for the project. Examples:
  • Literacy
  • Literature & Writing, Social Sciences
project_resource_summary An explanation of the resources needed for the project. Example:
  • My students need hands on literacy materials to manage sensory needs!
project_essay_1 First application essay*
project_essay_2 Second application essay*
project_essay_3 Third application essay*
project_essay_4 Fourth application essay*
project_submitted_datetime Datetime when project application was submitted. Example: 2016-04-28 12:43:56.245
teacher_id A unique identifier for the teacher of the proposed project. Example: bdf8baa8fedef6bfeec7ae4ff1c15c56
teacher_prefix Teacher's title. One of the following enumerated values:
  • nan
  • Dr.
  • Mr.
  • Mrs.
  • Ms.
  • Teacher.
teacher_number_of_previously_posted_projects Number of project applications previously submitted by the same teacher. Example: 2

* See the section Notes on the Essay Data for more details about these features.

Additionally, the resources.csv data set provides more data about the resources required for each project. Each line in this file represents a resource required by a project:

Feature Description
id A project_id value from the train.csv file. Example: p036502
description Desciption of the resource. Example: Tenor Saxophone Reeds, Box of 25
quantity Quantity of the resource required. Example: 3
price Price of the resource required. Example: 9.95

Note: Many projects require multiple resources. The id value corresponds to a project_id in train.csv, so you use it as a key to retrieve all resources needed for a project:

The data set contains the following label (the value you will attempt to predict):

Label Description
project_is_approved A binary flag indicating whether DonorsChoose approved the project. A value of 0 indicates the project was not approved, and a value of 1 indicates the project was approved.

Notes on the Essay Data

    Prior to May 17, 2016, the prompts for the essays were as follows:
  • __project_essay_1:__ "Introduce us to your classroom"
  • __project_essay_2:__ "Tell us more about your students"
  • __project_essay_3:__ "Describe how your students will use the materials you're requesting"
  • __project_essay_3:__ "Close by sharing why your project will make a difference"
    Starting on May 17, 2016, the number of essays was reduced from 4 to 2, and the prompts for the first 2 essays were changed to the following:
  • __project_essay_1:__ "Describe your students: What makes your students special? Specific details about their background, your neighborhood, and your school are all helpful."
  • __project_essay_2:__ "About your project: How will these materials make a difference in your students' learning and improve their school lives?"

  • For all projects with project_submitted_datetime of 2016-05-17 and later, the values of project_essay_3 and project_essay_4 will be NaN.
In [1]:
import os
os.chdir("F:/f/MY____/AAIC/AssignmentS/DonorsChoose_Data")
os.getcwd()
Out[1]:
'F:\\f\\MY____\\AAIC\\AssignmentS\\DonorsChoose_Data'
In [2]:
%matplotlib inline
import warnings
warnings.filterwarnings("ignore")

import sqlite3
import pandas as pd
import numpy as np
import nltk
import string
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.feature_extraction.text import TfidfTransformer
from sklearn.feature_extraction.text import TfidfVectorizer

from sklearn.feature_extraction.text import CountVectorizer
from sklearn.metrics import confusion_matrix
from sklearn import metrics
from sklearn.metrics import roc_curve, auc
from nltk.stem.porter import PorterStemmer

import re
# Tutorial about Python regular expressions: https://pymotw.com/2/re/
import string
from nltk.corpus import stopwords
from nltk.stem import PorterStemmer
from nltk.stem.wordnet import WordNetLemmatizer

from gensim.models import Word2Vec
from gensim.models import KeyedVectors
import pickle

from tqdm import tqdm
import os

from plotly import plotly
import plotly.offline as offline
import plotly.graph_objs as go
offline.init_notebook_mode()
from collections import Counter
C:\Users\Anvesh\Anaconda3\lib\site-packages\gensim\utils.py:1197: UserWarning: detected Windows; aliasing chunkize to chunkize_serial
  warnings.warn("detected Windows; aliasing chunkize to chunkize_serial")

1.1 Reading Data

In [3]:
from tqdm import tqdm
for i in tqdm(range(int(10e6))):
    pass
100%|█████████████████████████████████████████████████████████████████| 10000000/10000000 [00:04<00:00, 2043605.45it/s]
In [4]:
project_data = pd.read_csv('train_data.csv', nrows = 30000)
resource_data = pd.read_csv('resources.csv')
In [5]:
print("Number of data points in train data", project_data.shape)
print('-'*50)
print("The attributes of data :", project_data.columns.values)
Number of data points in train data (30000, 17)
--------------------------------------------------
The attributes of data : ['Unnamed: 0' 'id' 'teacher_id' 'teacher_prefix' 'school_state'
 'project_submitted_datetime' 'project_grade_category'
 'project_subject_categories' 'project_subject_subcategories'
 'project_title' 'project_essay_1' 'project_essay_2' 'project_essay_3'
 'project_essay_4' 'project_resource_summary'
 'teacher_number_of_previously_posted_projects' 'project_is_approved']
In [6]:
print("Number of data points in train data", resource_data.shape)
print(resource_data.columns.values)
resource_data.head(2)
Number of data points in train data (1541272, 4)
['id' 'description' 'quantity' 'price']
Out[6]:
id description quantity price
0 p233245 LC652 - Lakeshore Double-Space Mobile Drying Rack 1 149.00
1 p069063 Bouncy Bands for Desks (Blue support pipes) 3 14.95

1.2 preprocessing of project_subject_categories

In [7]:
catogories = list(project_data['project_subject_categories'].values)
# remove special characters from list of strings python: https://stackoverflow.com/a/47301924/4084039

# https://www.geeksforgeeks.org/removing-stop-words-nltk-python/
# https://stackoverflow.com/questions/23669024/how-to-strip-a-specific-word-from-a-string
# https://stackoverflow.com/questions/8270092/remove-all-whitespace-in-a-string-in-python
cat_list = []
for i in catogories:
    temp = ""
    # consider we have text like this "Math & Science, Warmth, Care & Hunger"
    for j in i.split(','): # it will split it in three parts ["Math & Science", "Warmth", "Care & Hunger"]
        if 'The' in j.split(): # this will split each of the catogory based on space "Math & Science"=> "Math","&", "Science"
            j=j.replace('The','') # if we have the words "The" we are going to replace it with ''(i.e removing 'The')
        j = j.replace(' ','') # we are placeing all the ' '(space) with ''(empty) ex:"Math & Science"=>"Math&Science"
        temp+=j.strip()+" " #" abc ".strip() will return "abc", remove the trailing spaces
        temp = temp.replace('&','_') # we are replacing the & value into 
    cat_list.append(temp.strip())
    
project_data['clean_categories'] = cat_list
project_data.drop(['project_subject_categories'], axis=1, inplace=True)

from collections import Counter
my_counter = Counter()
for word in project_data['clean_categories'].values:
    my_counter.update(word.split())

cat_dict = dict(my_counter)
sorted_cat_dict = dict(sorted(cat_dict.items(), key=lambda kv: kv[1]))

1.3 preprocessing of project_subject_subcategories

In [8]:
sub_catogories = list(project_data['project_subject_subcategories'].values)
# remove special characters from list of strings python: https://stackoverflow.com/a/47301924/4084039

# https://www.geeksforgeeks.org/removing-stop-words-nltk-python/
# https://stackoverflow.com/questions/23669024/how-to-strip-a-specific-word-from-a-string
# https://stackoverflow.com/questions/8270092/remove-all-whitespace-in-a-string-in-python

sub_cat_list = []
for i in sub_catogories:
    temp = ""
    # consider we have text like this "Math & Science, Warmth, Care & Hunger"
    for j in i.split(','): # it will split it in three parts ["Math & Science", "Warmth", "Care & Hunger"]
        if 'The' in j.split(): # this will split each of the catogory based on space "Math & Science"=> "Math","&", "Science"
            j=j.replace('The','') # if we have the words "The" we are going to replace it with ''(i.e removing 'The')
        j = j.replace(' ','') # we are placeing all the ' '(space) with ''(empty) ex:"Math & Science"=>"Math&Science"
        temp +=j.strip()+" "#" abc ".strip() will return "abc", remove the trailing spaces
        temp = temp.replace('&','_')
    sub_cat_list.append(temp.strip())

project_data['clean_subcategories'] = sub_cat_list
project_data.drop(['project_subject_subcategories'], axis=1, inplace=True)

# count of all the words in corpus python: https://stackoverflow.com/a/22898595/4084039
my_counter = Counter()
for word in project_data['clean_subcategories'].values:
    my_counter.update(word.split())
    
sub_cat_dict = dict(my_counter)
sorted_sub_cat_dict = dict(sorted(sub_cat_dict.items(), key=lambda kv: kv[1]))

1.3 Text preprocessing

In [9]:
# merge two column text dataframe: 
project_data["essay"] = project_data["project_essay_1"].map(str) +\
                        project_data["project_essay_2"].map(str) + \
                        project_data["project_essay_3"].map(str) + \
                        project_data["project_essay_4"].map(str)
In [10]:
project_data.head(2)
Out[10]:
Unnamed: 0 id teacher_id teacher_prefix school_state project_submitted_datetime project_grade_category project_title project_essay_1 project_essay_2 project_essay_3 project_essay_4 project_resource_summary teacher_number_of_previously_posted_projects project_is_approved clean_categories clean_subcategories essay
0 160221 p253737 c90749f5d961ff158d4b4d1e7dc665fc Mrs. IN 2016-12-05 13:43:57 Grades PreK-2 Educational Support for English Learners at Home My students are English learners that are work... \"The limits of your language are the limits o... NaN NaN My students need opportunities to practice beg... 0 0 Literacy_Language ESL Literacy My students are English learners that are work...
1 140945 p258326 897464ce9ddc600bced1151f324dd63a Mr. FL 2016-10-25 09:22:10 Grades 6-8 Wanted: Projector for Hungry Learners Our students arrive to our school eager to lea... The projector we need for our school is very c... NaN NaN My students need a projector to help with view... 7 1 History_Civics Health_Sports Civics_Government TeamSports Our students arrive to our school eager to lea...
In [11]:
# printing some random reviews
print(project_data['essay'].values[0])
print("="*50)
print(project_data['essay'].values[150])
print("="*50)
print(project_data['essay'].values[1000])
print("="*50)
print(project_data['essay'].values[20000])
print("="*50)
My students are English learners that are working on English as their second or third languages. We are a melting pot of refugees, immigrants, and native-born Americans bringing the gift of language to our school. \r\n\r\n We have over 24 languages represented in our English Learner program with students at every level of mastery.  We also have over 40 countries represented with the families within our school.  Each student brings a wealth of knowledge and experiences to us that open our eyes to new cultures, beliefs, and respect.\"The limits of your language are the limits of your world.\"-Ludwig Wittgenstein  Our English learner's have a strong support system at home that begs for more resources.  Many times our parents are learning to read and speak English along side of their children.  Sometimes this creates barriers for parents to be able to help their child learn phonetics, letter recognition, and other reading skills.\r\n\r\nBy providing these dvd's and players, students are able to continue their mastery of the English language even if no one at home is able to assist.  All families with students within the Level 1 proficiency status, will be a offered to be a part of this program.  These educational videos will be specially chosen by the English Learner Teacher and will be sent home regularly to watch.  The videos are to help the child develop early reading skills.\r\n\r\nParents that do not have access to a dvd player will have the opportunity to check out a dvd player to use for the year.  The plan is to use these videos and educational dvd's for the years to come for other EL students.\r\nnannan
==================================================
The 51 fifth grade students that will cycle through my classroom this year all love learning, at least most of the time. At our school, 97.3% of the students receive free or reduced price lunch. Of the 560 students, 97.3% are minority students. \r\nThe school has a vibrant community that loves to get together and celebrate. Around Halloween there is a whole school parade to show off the beautiful costumes that students wear. On Cinco de Mayo we put on a big festival with crafts made by the students, dances, and games. At the end of the year the school hosts a carnival to celebrate the hard work put in during the school year, with a dunk tank being the most popular activity.My students will use these five brightly colored Hokki stools in place of regular, stationary, 4-legged chairs. As I will only have a total of ten in the classroom and not enough for each student to have an individual one, they will be used in a variety of ways. During independent reading time they will be used as special chairs students will each use on occasion. I will utilize them in place of chairs at my small group tables during math and reading times. The rest of the day they will be used by the students who need the highest amount of movement in their life in order to stay focused on school.\r\n\r\nWhenever asked what the classroom is missing, my students always say more Hokki Stools. They can't get their fill of the 5 stools we already have. When the students are sitting in group with me on the Hokki Stools, they are always moving, but at the same time doing their work. Anytime the students get to pick where they can sit, the Hokki Stools are the first to be taken. There are always students who head over to the kidney table to get one of the stools who are disappointed as there are not enough of them. \r\n\r\nWe ask a lot of students to sit for 7 hours a day. The Hokki stools will be a compromise that allow my students to do desk work and move at the same time. These stools will help students to meet their 60 minutes a day of movement by allowing them to activate their core muscles for balance while they sit. For many of my students, these chairs will take away the barrier that exists in schools for a child who can't sit still.nannan
==================================================
How do you remember your days of school? Was it in a sterile environment with plain walls, rows of desks, and a teacher in front of the room? A typical day in our room is nothing like that. I work hard to create a warm inviting themed room for my students look forward to coming to each day.\r\n\r\nMy class is made up of 28 wonderfully unique boys and girls of mixed races in Arkansas.\r\nThey attend a Title I school, which means there is a high enough percentage of free and reduced-price lunch to qualify. Our school is an \"open classroom\" concept, which is very unique as there are no walls separating the classrooms. These 9 and 10 year-old students are very eager learners; they are like sponges, absorbing all the information and experiences and keep on wanting more.With these resources such as the comfy red throw pillows and the whimsical nautical hanging decor and the blue fish nets, I will be able to help create the mood in our classroom setting to be one of a themed nautical environment. Creating a classroom environment is very important in the success in each and every child's education. The nautical photo props will be used with each child as they step foot into our classroom for the first time on Meet the Teacher evening. I'll take pictures of each child with them, have them developed, and then hung in our classroom ready for their first day of 4th grade.  This kind gesture will set the tone before even the first day of school! The nautical thank you cards will be used throughout the year by the students as they create thank you cards to their team groups.\r\n\r\nYour generous donations will help me to help make our classroom a fun, inviting, learning environment from day one.\r\n\r\nIt costs lost of money out of my own pocket on resources to get our classroom ready. Please consider helping with this project to make our new school year a very successful one. Thank you!nannan
==================================================
My kindergarten students have varied disabilities ranging from speech and language delays, cognitive delays, gross/fine motor delays, to autism. They are eager beavers and always strive to work their hardest working past their limitations. \r\n\r\nThe materials we have are the ones I seek out for my students. I teach in a Title I school where most of the students receive free or reduced price lunch.  Despite their disabilities and limitations, my students love coming to school and come eager to learn and explore.Have you ever felt like you had ants in your pants and you needed to groove and move as you were in a meeting? This is how my kids feel all the time. The want to be able to move as they learn or so they say.Wobble chairs are the answer and I love then because they develop their core, which enhances gross motor and in Turn fine motor skills. \r\nThey also want to learn through games, my kids don't want to sit and do worksheets. They want to learn to count by jumping and playing. Physical engagement is the key to our success. The number toss and color and shape mats can make that happen. My students will forget they are doing work and just have the fun a 6 year old deserves.nannan
==================================================

Introducing New Features

Consider these set of features for Set 5 in Assignment:

categorical dataschool_state clean_categories....clean_subcategories....project_grade_category....teacher_prefix

numerical data quantity....teacher_number_of_previously_posted_projects....price

New Features

sentiment score's of each of the essay : numerical data
number of words in the title : numerical data
number of words in the combine essays : numerical data

In [12]:
import re
def decontracted(phrase):
    # specific
    phrase = re.sub(r"won't", "will not", phrase)
    phrase = re.sub(r"can\'t", "can not", phrase)

    # general
    phrase = re.sub(r"n\'t", " not", phrase)
    phrase = re.sub(r"\'re", " are", phrase)
    phrase = re.sub(r"\'s", " is", phrase)
    phrase = re.sub(r"\'d", " would", phrase)
    phrase = re.sub(r"\'ll", " will", phrase)
    phrase = re.sub(r"\'t", " not", phrase)
    phrase = re.sub(r"\'ve", " have", phrase)
    phrase = re.sub(r"\'m", " am", phrase)
    return phrase
In [13]:
new_title = []
for i in tqdm(project_data['project_title']):
    j = decontracted(i)
    new_title.append(j)   
    
100%|█████████████████████████████████████████████████████████████████████████| 30000/30000 [00:00<00:00, 65235.75it/s]
In [14]:
#Introducing New Features
title_word_count = []
#for i in project_data['project_title']:
for i in tqdm(new_title):
    j = len(i.split())
    title_word_count.append(j)
    #print(j)
project_data['title_word_count'] = title_word_count
100%|████████████████████████████████████████████████████████████████████████| 30000/30000 [00:00<00:00, 447839.87it/s]
In [15]:
project_data.head(2)
Out[15]:
Unnamed: 0 id teacher_id teacher_prefix school_state project_submitted_datetime project_grade_category project_title project_essay_1 project_essay_2 project_essay_3 project_essay_4 project_resource_summary teacher_number_of_previously_posted_projects project_is_approved clean_categories clean_subcategories essay title_word_count
0 160221 p253737 c90749f5d961ff158d4b4d1e7dc665fc Mrs. IN 2016-12-05 13:43:57 Grades PreK-2 Educational Support for English Learners at Home My students are English learners that are work... \"The limits of your language are the limits o... NaN NaN My students need opportunities to practice beg... 0 0 Literacy_Language ESL Literacy My students are English learners that are work... 7
1 140945 p258326 897464ce9ddc600bced1151f324dd63a Mr. FL 2016-10-25 09:22:10 Grades 6-8 Wanted: Projector for Hungry Learners Our students arrive to our school eager to lea... The projector we need for our school is very c... NaN NaN My students need a projector to help with view... 7 1 History_Civics Health_Sports Civics_Government TeamSports Our students arrive to our school eager to lea... 5
In [16]:
new_essay = []
for i in tqdm(project_data['essay']):
    j = decontracted(i)
    new_essay.append(j)  
100%|█████████████████████████████████████████████████████████████████████████| 30000/30000 [00:00<00:00, 38190.81it/s]
In [17]:
essay_word_count = []
for i in tqdm(new_essay):
    j = len(i.split())
    essay_word_count.append(j)
    #print(j)
project_data['essay_word_count'] = essay_word_count
100%|█████████████████████████████████████████████████████████████████████████| 30000/30000 [00:00<00:00, 39464.88it/s]
In [18]:
project_data.head(2)
Out[18]:
Unnamed: 0 id teacher_id teacher_prefix school_state project_submitted_datetime project_grade_category project_title project_essay_1 project_essay_2 project_essay_3 project_essay_4 project_resource_summary teacher_number_of_previously_posted_projects project_is_approved clean_categories clean_subcategories essay title_word_count essay_word_count
0 160221 p253737 c90749f5d961ff158d4b4d1e7dc665fc Mrs. IN 2016-12-05 13:43:57 Grades PreK-2 Educational Support for English Learners at Home My students are English learners that are work... \"The limits of your language are the limits o... NaN NaN My students need opportunities to practice beg... 0 0 Literacy_Language ESL Literacy My students are English learners that are work... 7 275
1 140945 p258326 897464ce9ddc600bced1151f324dd63a Mr. FL 2016-10-25 09:22:10 Grades 6-8 Wanted: Projector for Hungry Learners Our students arrive to our school eager to lea... The projector we need for our school is very c... NaN NaN My students need a projector to help with view... 7 1 History_Civics Health_Sports Civics_Government TeamSports Our students arrive to our school eager to lea... 5 221

Computing Sentiment Scores

In [19]:
import nltk
from nltk.sentiment.vader import SentimentIntensityAnalyzer

# import nltk
#nltk.download('vader_lexicon')

sid = SentimentIntensityAnalyzer()

for_sentiment = 'a person is a person no matter how small dr seuss i teach the smallest students with the biggest enthusiasm'
ss = sid.polarity_scores(for_sentiment)

for k in ss:
    print('{0}: {1}, '.format(k, ss[k]), end='')

# we can use these 4 things as features/attributes (neg, neu, pos, compound)
# neg: 0.0, neu: 0.753, pos: 0.247, compound: 0.93
neg: 0.109, neu: 0.693, pos: 0.198, compound: 0.2023, 
In [20]:
SID = SentimentIntensityAnalyzer()
#There is NEGITIVE and POSITIVE and NEUTRAL and COMPUND SCORES
#http://www.nltk.org/howto/sentiment.html
    
negitive = []
positive = []
neutral = []
compound = []
for i in tqdm(project_data['essay']):
    j = SID.polarity_scores(i)['neg']
    k = SID.polarity_scores(i)['neu']
    l = SID.polarity_scores(i)['pos']
    m = SID.polarity_scores(i)['compound']
    negitive.append(j)
    positive.append(k)
    neutral.append(l)
    compound.append(m)
100%|████████████████████████████████████████████████████████████████████████████| 30000/30000 [08:32<00:00, 58.56it/s]
In [21]:
project_data['negitive'] = negitive
project_data['positive'] = positive
project_data['neutral'] = neutral
project_data['compound'] = compound
In [22]:
project_data.head(2)
Out[22]:
Unnamed: 0 id teacher_id teacher_prefix school_state project_submitted_datetime project_grade_category project_title project_essay_1 project_essay_2 ... project_is_approved clean_categories clean_subcategories essay title_word_count essay_word_count negitive positive neutral compound
0 160221 p253737 c90749f5d961ff158d4b4d1e7dc665fc Mrs. IN 2016-12-05 13:43:57 Grades PreK-2 Educational Support for English Learners at Home My students are English learners that are work... \"The limits of your language are the limits o... ... 0 Literacy_Language ESL Literacy My students are English learners that are work... 7 275 0.008 0.911 0.081 0.9611
1 140945 p258326 897464ce9ddc600bced1151f324dd63a Mr. FL 2016-10-25 09:22:10 Grades 6-8 Wanted: Projector for Hungry Learners Our students arrive to our school eager to lea... The projector we need for our school is very c... ... 1 History_Civics Health_Sports Civics_Government TeamSports Our students arrive to our school eager to lea... 5 221 0.037 0.851 0.112 0.9267

2 rows × 24 columns

Preparing Data for Models

In [23]:
# https://stackoverflow.com/a/47091490/4084039
import re

def decontracted(phrase):
    # specific
    phrase = re.sub(r"won't", "will not", phrase)
    phrase = re.sub(r"can\'t", "can not", phrase)

    # general
    phrase = re.sub(r"n\'t", " not", phrase)
    phrase = re.sub(r"\'re", " are", phrase)
    phrase = re.sub(r"\'s", " is", phrase)
    phrase = re.sub(r"\'d", " would", phrase)
    phrase = re.sub(r"\'ll", " will", phrase)
    phrase = re.sub(r"\'t", " not", phrase)
    phrase = re.sub(r"\'ve", " have", phrase)
    phrase = re.sub(r"\'m", " am", phrase)
    return phrase
In [24]:
sent = decontracted(project_data['essay'].values[20000])
print(sent)
print("="*50)
My kindergarten students have varied disabilities ranging from speech and language delays, cognitive delays, gross/fine motor delays, to autism. They are eager beavers and always strive to work their hardest working past their limitations. \r\n\r\nThe materials we have are the ones I seek out for my students. I teach in a Title I school where most of the students receive free or reduced price lunch.  Despite their disabilities and limitations, my students love coming to school and come eager to learn and explore.Have you ever felt like you had ants in your pants and you needed to groove and move as you were in a meeting? This is how my kids feel all the time. The want to be able to move as they learn or so they say.Wobble chairs are the answer and I love then because they develop their core, which enhances gross motor and in Turn fine motor skills. \r\nThey also want to learn through games, my kids do not want to sit and do worksheets. They want to learn to count by jumping and playing. Physical engagement is the key to our success. The number toss and color and shape mats can make that happen. My students will forget they are doing work and just have the fun a 6 year old deserves.nannan
==================================================
In [25]:
# \r \n \t remove from string python: http://texthandler.com/info/remove-line-breaks-python/
sent = sent.replace('\\r', ' ')
sent = sent.replace('\\"', ' ')
sent = sent.replace('\\n', ' ')
print(sent)
My kindergarten students have varied disabilities ranging from speech and language delays, cognitive delays, gross/fine motor delays, to autism. They are eager beavers and always strive to work their hardest working past their limitations.     The materials we have are the ones I seek out for my students. I teach in a Title I school where most of the students receive free or reduced price lunch.  Despite their disabilities and limitations, my students love coming to school and come eager to learn and explore.Have you ever felt like you had ants in your pants and you needed to groove and move as you were in a meeting? This is how my kids feel all the time. The want to be able to move as they learn or so they say.Wobble chairs are the answer and I love then because they develop their core, which enhances gross motor and in Turn fine motor skills.   They also want to learn through games, my kids do not want to sit and do worksheets. They want to learn to count by jumping and playing. Physical engagement is the key to our success. The number toss and color and shape mats can make that happen. My students will forget they are doing work and just have the fun a 6 year old deserves.nannan
In [26]:
#remove spacial character: https://stackoverflow.com/a/5843547/4084039
sent = re.sub('[^A-Za-z0-9]+', ' ', sent)
print(sent)
My kindergarten students have varied disabilities ranging from speech and language delays cognitive delays gross fine motor delays to autism They are eager beavers and always strive to work their hardest working past their limitations The materials we have are the ones I seek out for my students I teach in a Title I school where most of the students receive free or reduced price lunch Despite their disabilities and limitations my students love coming to school and come eager to learn and explore Have you ever felt like you had ants in your pants and you needed to groove and move as you were in a meeting This is how my kids feel all the time The want to be able to move as they learn or so they say Wobble chairs are the answer and I love then because they develop their core which enhances gross motor and in Turn fine motor skills They also want to learn through games my kids do not want to sit and do worksheets They want to learn to count by jumping and playing Physical engagement is the key to our success The number toss and color and shape mats can make that happen My students will forget they are doing work and just have the fun a 6 year old deserves nannan
In [27]:
# https://gist.github.com/sebleier/554280
# we are removing the words from the stop words list: 'no', 'nor', 'not'
stopwords= ['i', 'me', 'my', 'myself', 'we', 'our', 'ours', 'ourselves', 'you', "you're", "you've",\
            "you'll", "you'd", 'your', 'yours', 'yourself', 'yourselves', 'he', 'him', 'his', 'himself', \
            'she', "she's", 'her', 'hers', 'herself', 'it', "it's", 'its', 'itself', 'they', 'them', 'their',\
            'theirs', 'themselves', 'what', 'which', 'who', 'whom', 'this', 'that', "that'll", 'these', 'those', \
            'am', 'is', 'are', 'was', 'were', 'be', 'been', 'being', 'have', 'has', 'had', 'having', 'do', 'does', \
            'did', 'doing', 'a', 'an', 'the', 'and', 'but', 'if', 'or', 'because', 'as', 'until', 'while', 'of', \
            'at', 'by', 'for', 'with', 'about', 'against', 'between', 'into', 'through', 'during', 'before', 'after',\
            'above', 'below', 'to', 'from', 'up', 'down', 'in', 'out', 'on', 'off', 'over', 'under', 'again', 'further',\
            'then', 'once', 'here', 'there', 'when', 'where', 'why', 'how', 'all', 'any', 'both', 'each', 'few', 'more',\
            'most', 'other', 'some', 'such', 'only', 'own', 'same', 'so', 'than', 'too', 'very', \
            's', 't', 'can', 'will', 'just', 'don', "don't", 'should', "should've", 'now', 'd', 'll', 'm', 'o', 're', \
            've', 'y', 'ain', 'aren', "aren't", 'couldn', "couldn't", 'didn', "didn't", 'doesn', "doesn't", 'hadn',\
            "hadn't", 'hasn', "hasn't", 'haven', "haven't", 'isn', "isn't", 'ma', 'mightn', "mightn't", 'mustn',\
            "mustn't", 'needn', "needn't", 'shan', "shan't", 'shouldn', "shouldn't", 'wasn', "wasn't", 'weren', "weren't", \
            'won', "won't", 'wouldn', "wouldn't"]

Project_Essays preprocessing

In [28]:
#train_preprocessed_essays 
# Combining all the above stundents 
from tqdm import tqdm
preprocessed_essays = []
# tqdm is for printing the status bar
for sentance in tqdm(project_data['essay'].values):
    sent = decontracted(sentance)
    sent = sent.replace('\\r', ' ')
    sent = sent.replace('\\"', ' ')
    sent = sent.replace('\\n', ' ')
    sent = re.sub('[^A-Za-z0-9]+', ' ', sent)
    # https://gist.github.com/sebleier/554280
    sent = ' '.join(e for e in sent.split() if e not in stopwords)
    preprocessed_essays.append(sent.lower().strip())
100%|██████████████████████████████████████████████████████████████████████████| 30000/30000 [00:24<00:00, 1227.62it/s]
In [29]:
preprocessed_essays[10]
Out[29]:
'there many little ways enlarge world love books best jacqueline kennedy i work title 1 school serves students lower income families they often enter 5th grade reading second grade level notion reading another thing not good i provide motivation want pick book snuggle bean bag read ever lifetime i 8 sets students enjoy books classroom library throughout past years result falling apart it testament much loved reading unfortunately books soon need retired with purchase new books ensure joy reading continues past school year a haunted mystery series word mouse the babysitters club books students would love read my students struggle building vocabulary time reach 5th grade no interest picking book i determined change futures sharing joy reading reading allows make connections beyond limited world gives gateway limitless opportunities your donations help improve classroom library tangible evidence students i not one thinks reading changes lives nannan'

Project_Titles_preprocessing

In [30]:
# train_preprocessed_title
from tqdm import tqdm
preprocessed_titles = []
# tqdm is for printing the status bar
for sentance in tqdm(project_data['project_title'].values):
    sent = decontracted(sentance)
    sent = sent.replace('\\r', ' ')
    sent = sent.replace('\\"', ' ')
    sent = sent.replace('\\n', ' ')
    sent = re.sub('[^A-Za-z0-9]+', ' ', sent)
    # https://gist.github.com/sebleier/554280
    sent = ' '.join(e for e in sent.split() if e not in stopwords)
    preprocessed_titles.append(sent.lower().strip())
100%|█████████████████████████████████████████████████████████████████████████| 30000/30000 [00:01<00:00, 25942.42it/s]
In [31]:
preprocessed_titles[10]
Out[31]:
'reading changes lives'
In [32]:
project_data.columns
Out[32]:
Index(['Unnamed: 0', 'id', 'teacher_id', 'teacher_prefix', 'school_state',
       'project_submitted_datetime', 'project_grade_category', 'project_title',
       'project_essay_1', 'project_essay_2', 'project_essay_3',
       'project_essay_4', 'project_resource_summary',
       'teacher_number_of_previously_posted_projects', 'project_is_approved',
       'clean_categories', 'clean_subcategories', 'essay', 'title_word_count',
       'essay_word_count', 'negitive', 'positive', 'neutral', 'compound'],
      dtype='object')

we are going to consider

   - school_state : categorical data
   - clean_categories : categorical data
   - clean_subcategories : categorical data
   - project_grade_category : categorical data
   - teacher_prefix : categorical data

   - project_title : text data
   - text : text data
   - project_resource_summary: text data (optinal)

   - quantity : numerical (optinal)
   - teacher_number_of_previously_posted_projects : numerical
   - price : numerical
In [33]:
project_data.drop(['essay'], axis=1, inplace=True)
project_data.drop(['project_title'], axis=1, inplace=True)
In [34]:
project_data['clean_essay'] = preprocessed_essays
project_data['clean_project_title'] = preprocessed_titles
project_data.head(1)
Out[34]:
Unnamed: 0 id teacher_id teacher_prefix school_state project_submitted_datetime project_grade_category project_essay_1 project_essay_2 project_essay_3 ... clean_categories clean_subcategories title_word_count essay_word_count negitive positive neutral compound clean_essay clean_project_title
0 160221 p253737 c90749f5d961ff158d4b4d1e7dc665fc Mrs. IN 2016-12-05 13:43:57 Grades PreK-2 My students are English learners that are work... \"The limits of your language are the limits o... NaN ... Literacy_Language ESL Literacy 7 275 0.008 0.911 0.081 0.9611 my students english learners working english s... educational support english learners home

1 rows × 24 columns

In [35]:
project_data = project_data.fillna(project_data['teacher_prefix'].value_counts().index[0])
In [36]:
#Train Test Split 
from sklearn.model_selection import train_test_split
x_train, x_test, y_train, y_test = train_test_split(project_data, project_data["project_is_approved"],
                                                    test_size = 0.33, stratify = project_data["project_is_approved"],
                                                    random_state = 42)
#Train CV Split
#x_train, x_cv, y_train, y_cv = train_test_split(x_train, y_train, test_size = 0.33, stratify = y_train,
#                                                random_state = 42)
In [37]:
print(x_test.columns)
print(x_train.columns)
Index(['Unnamed: 0', 'id', 'teacher_id', 'teacher_prefix', 'school_state',
       'project_submitted_datetime', 'project_grade_category',
       'project_essay_1', 'project_essay_2', 'project_essay_3',
       'project_essay_4', 'project_resource_summary',
       'teacher_number_of_previously_posted_projects', 'project_is_approved',
       'clean_categories', 'clean_subcategories', 'title_word_count',
       'essay_word_count', 'negitive', 'positive', 'neutral', 'compound',
       'clean_essay', 'clean_project_title'],
      dtype='object')
Index(['Unnamed: 0', 'id', 'teacher_id', 'teacher_prefix', 'school_state',
       'project_submitted_datetime', 'project_grade_category',
       'project_essay_1', 'project_essay_2', 'project_essay_3',
       'project_essay_4', 'project_resource_summary',
       'teacher_number_of_previously_posted_projects', 'project_is_approved',
       'clean_categories', 'clean_subcategories', 'title_word_count',
       'essay_word_count', 'negitive', 'positive', 'neutral', 'compound',
       'clean_essay', 'clean_project_title'],
      dtype='object')

1.5.1 Vectorizing Categorical data Using Response Coding

In [38]:
def Responsetable(table, col) :
    cat = table[col].unique()
    
    freq_Pos = []
    for i in cat : 
        freq_Pos.append(len(table.loc[(table[col] == i) & (table['project_is_approved'] == 1)]))
        
        
    freq_Neg = []
    for i in cat : 
        freq_Neg.append(len(table.loc[(table[col] == i) & (table['project_is_approved'] == 0)]))
        
    
    encoded_Pos = []
    for i in range(len(cat)) :
        encoded_Pos.append(freq_Pos[i]/(freq_Pos[i] + freq_Neg[i]))
    
    encoded_Neg = []
    encoded_Neg[:] = [1 - x for x in encoded_Pos]
        
    encoded_Pos_val = dict(zip(cat, encoded_Pos)) 
    encoded_Neg_val = dict(zip(cat, encoded_Neg)) 
    
    return encoded_Pos_val, encoded_Neg_val
In [39]:
def Responsecode(table) : 
    pos_cleancat, neg_cleancat = Responsetable(table,'clean_categories')
    pos_cleansubcat, neg_cleansubcat = Responsetable(table,'clean_subcategories')
    pos_schoolstate, neg_schoolstate = Responsetable(table, 'school_state')
    pos_teacherprefix, neg_teacherprefix = Responsetable(table, 'teacher_prefix')
    pos_projgradecat, neg_projgradecat = Responsetable(table, 'project_grade_category')
    
    df = pd.DataFrame()
    df['clean_cat_pos'] = table['clean_categories'].map(pos_cleancat)
    df['clean_cat_neg'] = table['clean_categories'].map(neg_cleancat)
    df['clean_subcat_pos'] = table['clean_subcategories'].map(pos_cleansubcat)
    df['clean_subcat_neg'] = table['clean_subcategories'].map(neg_cleansubcat)
    df['school_state_pos'] = table['school_state'].map(pos_schoolstate)
    df['school_state_neg'] = table['school_state'].map(neg_schoolstate)
    df['teacher_prefix_pos'] = table['teacher_prefix'].map(pos_teacherprefix)
    df['teacher_prefix_neg'] = table['teacher_prefix'].map(neg_teacherprefix)
    df['proj_grade_cat_pos'] = table['project_grade_category'].map(pos_projgradecat)
    df['proj_grade_cat_neg'] = table['project_grade_category'].map(neg_projgradecat)

    return df
In [40]:
newTrain = Responsecode(x_train)
newTest = Responsecode(x_test)
In [41]:
def mergeEncoding(table, p, n) :
    lstPos = table[p].values.tolist()
    lstNeg = table[n].values.tolist()
    frame = pd.DataFrame(list(zip(lstNeg, lstPos)))
    
    return frame
In [42]:
#Clean Categories
X_train_clean_cat_ohe = mergeEncoding(newTrain, 'clean_cat_pos', 'clean_cat_neg')
X_test_clean_cat_ohe = mergeEncoding(newTest, 'clean_cat_pos', 'clean_cat_neg')
In [43]:
print(X_train_clean_cat_ohe.shape)
X_test_clean_cat_ohe.shape
(20100, 2)
Out[43]:
(9900, 2)
In [44]:
#Clean SUB Categories
X_train_clean_subcat_ohe = mergeEncoding(newTrain, 'clean_subcat_pos', 'clean_subcat_neg')
X_test_clean_subcat_ohe = mergeEncoding(newTest, 'clean_subcat_pos', 'clean_subcat_neg')
print(X_train_clean_subcat_ohe.shape)
print(X_test_clean_subcat_ohe.shape)
(20100, 2)
(9900, 2)
In [45]:
#Project Grade Category
X_train_grade_ohe = mergeEncoding(newTrain, 'proj_grade_cat_pos', 'proj_grade_cat_neg')
X_test_grade_ohe = mergeEncoding(newTest, 'proj_grade_cat_pos', 'proj_grade_cat_neg')
print(X_train_grade_ohe.shape)
print(X_test_grade_ohe.shape)
(20100, 2)
(9900, 2)
In [46]:
#School State
X_train_state_ohe = mergeEncoding(newTrain, 'school_state_pos', 'school_state_neg')
X_test_state_ohe = mergeEncoding(newTest, 'school_state_pos', 'school_state_neg')
print(X_train_state_ohe.shape)
print(X_test_state_ohe.shape)
(20100, 2)
(9900, 2)
In [47]:
#Teacher Prefix
X_train_teacher_ohe = mergeEncoding(newTrain, 'teacher_prefix_pos', 'teacher_prefix_neg')
X_test_teacher_ohe = mergeEncoding(newTest, 'teacher_prefix_pos', 'teacher_prefix_neg')
print(X_train_teacher_ohe.shape)
print(X_test_teacher_ohe.shape)
(20100, 2)
(9900, 2)

1.5.2 Vectorizing Text data

1.5.2.1 Bag of words

Minimum frequency of words 10 (min_df = 10) of ALL features

In [48]:
from sklearn.feature_extraction.text import CountVectorizer
# We are considering only the words which appeared in at least 10 documents(rows or projects).
vectorizer = CountVectorizer(min_df=10)
vectorizer.fit(x_train['clean_essay'])

train_text_bow = vectorizer.fit_transform(x_train['clean_essay'])
print("Shape of matrix after one hot encodig ",train_text_bow.shape, y_train.shape)
Shape of matrix after one hot encodig  (20100, 8528) (20100,)
In [49]:
#Vectorizing Test Data
# you can vectorize the title also 
# before you vectorize the title make sure you preprocess it
test_text_bow = vectorizer.transform(x_test['clean_essay'])
print("Shape of matrix after one hot encodig ",test_text_bow.shape, y_test.shape)
Shape of matrix after one hot encodig  (9900, 8528) (9900,)

Project_Title BOW

In [50]:
# We are considering only the words which appeared in at least 10 documents(rows or projects).
vectorizer = CountVectorizer(min_df=10)
vectorizer.fit(x_train['clean_project_title'])

train_title_bow = vectorizer.fit_transform(x_train['clean_project_title'])
print("Shape of matrix after one hot encodig ",train_title_bow.shape)
Shape of matrix after one hot encodig  (20100, 1153)
In [51]:
#Vectorizing Test Data
test_title_bow = vectorizer.transform(x_test['clean_project_title'])
print("Shape of matrix after one hot encodig ",test_title_bow.shape)
Shape of matrix after one hot encodig  (9900, 1153)

1.5.2.2 TFIDF vectorizer

Here also we are considering word frequency of 10 words(min_df = 10) of all features

In [52]:
from sklearn.feature_extraction.text import TfidfVectorizer
vectorizer = TfidfVectorizer(min_df=10)
vectorizer.fit(x_train['clean_essay'])

train_text_tfidf = vectorizer.fit_transform(x_train['clean_essay'])
print("Shape of matrix after one hot encodig ",train_text_tfidf.shape)
Shape of matrix after one hot encodig  (20100, 8528)
In [53]:
test_text_tfidf = vectorizer.transform(x_test['clean_essay'])
print("Shape of matrix after one hot encodig ",test_text_tfidf.shape)
Shape of matrix after one hot encodig  (9900, 8528)

Project Titles

In [54]:
vectorizer = TfidfVectorizer(min_df=10)
vectorizer.fit(x_train['clean_project_title'])

train_title_tfidf = vectorizer.fit_transform(x_train['clean_project_title'])
print("Shape of matrix after one hot encodig ",train_title_tfidf.shape)
Shape of matrix after one hot encodig  (20100, 1153)
In [55]:
test_title_tfidf = vectorizer.transform(x_test['clean_project_title'])
print("Shape of matrix after one hot encodig ",test_title_tfidf.shape)
Shape of matrix after one hot encodig  (9900, 1153)

1.5.2.3 Using Pretrained Models: Avg W2V

In [56]:
'''
# Reading glove vectors in python: https://stackoverflow.com/a/38230349/4084039
def loadGloveModel(gloveFile):
    print ("Loading Glove Model")
    f = open(gloveFile,'r', encoding="utf8")
    model = {}
    for line in tqdm(f):
        splitLine = line.split()
        word = splitLine[0]
        embedding = np.array([float(val) for val in splitLine[1:]])
        model[word] = embedding
    print ("Done.",len(model)," words loaded!")
    return model
model = loadGloveModel('glove.42B.300d.txt')

# ============================
Output:
    
Loading Glove Model
1917495it [06:32, 4879.69it/s]
Done. 1917495  words loaded!

# ============================

words = []
for i in preproced_texts:
    words.extend(i.split(' '))

for i in preproced_titles:
    words.extend(i.split(' '))
print("all the words in the coupus", len(words))
words = set(words)
print("the unique words in the coupus", len(words))

inter_words = set(model.keys()).intersection(words)
print("The number of words that are present in both glove vectors and our coupus", \
      len(inter_words),"(",np.round(len(inter_words)/len(words)*100,3),"%)")

words_courpus = {}
words_glove = set(model.keys())
for i in words:
    if i in words_glove:
        words_courpus[i] = model[i]
print("word 2 vec length", len(words_courpus))


# stronging variables into pickle files python: http://www.jessicayung.com/how-to-use-pickle-to-save-and-load-variables-in-python/

import pickle
with open('glove_vectors', 'wb') as f:
    pickle.dump(words_courpus, f)


'''
Out[56]:
'\n# Reading glove vectors in python: https://stackoverflow.com/a/38230349/4084039\ndef loadGloveModel(gloveFile):\n    print ("Loading Glove Model")\n    f = open(gloveFile,\'r\', encoding="utf8")\n    model = {}\n    for line in tqdm(f):\n        splitLine = line.split()\n        word = splitLine[0]\n        embedding = np.array([float(val) for val in splitLine[1:]])\n        model[word] = embedding\n    print ("Done.",len(model)," words loaded!")\n    return model\nmodel = loadGloveModel(\'glove.42B.300d.txt\')\n\n# ============================\nOutput:\n    \nLoading Glove Model\n1917495it [06:32, 4879.69it/s]\nDone. 1917495  words loaded!\n\n# ============================\n\nwords = []\nfor i in preproced_texts:\n    words.extend(i.split(\' \'))\n\nfor i in preproced_titles:\n    words.extend(i.split(\' \'))\nprint("all the words in the coupus", len(words))\nwords = set(words)\nprint("the unique words in the coupus", len(words))\n\ninter_words = set(model.keys()).intersection(words)\nprint("The number of words that are present in both glove vectors and our coupus",       len(inter_words),"(",np.round(len(inter_words)/len(words)*100,3),"%)")\n\nwords_courpus = {}\nwords_glove = set(model.keys())\nfor i in words:\n    if i in words_glove:\n        words_courpus[i] = model[i]\nprint("word 2 vec length", len(words_courpus))\n\n\n# stronging variables into pickle files python: http://www.jessicayung.com/how-to-use-pickle-to-save-and-load-variables-in-python/\n\nimport pickle\nwith open(\'glove_vectors\', \'wb\') as f:\n    pickle.dump(words_courpus, f)\n\n\n'
In [57]:
# stronging variables into pickle files python: http://www.jessicayung.com/how-to-use-pickle-to-save-and-load-variables-in-python/
# make sure you have the glove_vectors file
with open('glove_vectors', 'rb') as f:
    model = pickle.load(f)
    glove_words =  set(model.keys())
In [58]:
# average Word2Vec
# compute average word2vec for each review.
train_avg_w2v_vectors = []; # the avg-w2v for each sentence/review is stored in this list
for sentence in tqdm(x_train['clean_essay']): # for each review/sentence
    vector = np.zeros(300) # as word vectors are of zero length
    cnt_words =0; # num of words with a valid vector in the sentence/review
    for word in sentence.split(): # for each word in a review/sentence
        if word in glove_words:
            vector += model[word]
            cnt_words += 1
    if cnt_words != 0:
        vector /= cnt_words
    train_avg_w2v_vectors.append(vector)

print(len(train_avg_w2v_vectors))
print(len(train_avg_w2v_vectors[0]))
100%|██████████████████████████████████████████████████████████████████████████| 20100/20100 [00:08<00:00, 2255.29it/s]
20100
300
In [59]:
# average Word2Vec
# compute average word2vec for each review.
test_avg_w2v_vectors = []; # the avg-w2v for each sentence/review is stored in this list
for sentence in tqdm(x_test['clean_essay']): # for each review/sentence
    vector = np.zeros(300) # as word vectors are of zero length
    cnt_words =0; # num of words with a valid vector in the sentence/review
    for word in sentence.split(): # for each word in a review/sentence
        if word in glove_words:
            vector += model[word]
            cnt_words += 1
    if cnt_words != 0:
        vector /= cnt_words
    test_avg_w2v_vectors.append(vector)

print(len(test_avg_w2v_vectors))
print(len(test_avg_w2v_vectors[0]))
100%|████████████████████████████████████████████████████████████████████████████| 9900/9900 [00:04<00:00, 2308.31it/s]
9900
300

AVG_W2V Project_Titles

In [60]:
# average Word2Vec
# compute average word2vec for each review.
train_title_avg_w2v_vectors = []; # the avg-w2v for each sentence/review is stored in this list
for sentence in tqdm(x_train['clean_project_title']): # for each review/sentence
    vector = np.zeros(300) # as word vectors are of zero length
    cnt_words =0; # num of words with a valid vector in the sentence/review
    for word in sentence.split(): # for each word in a review/sentence
        if word in glove_words:
            vector += model[word]
            cnt_words += 1
    if cnt_words != 0:
        vector /= cnt_words
    train_title_avg_w2v_vectors.append(vector)

print(len(train_title_avg_w2v_vectors))
print(len(train_title_avg_w2v_vectors[0]))
100%|█████████████████████████████████████████████████████████████████████████| 20100/20100 [00:00<00:00, 40493.51it/s]
20100
300
In [61]:
# average Word2Vec
# compute average word2vec for each review.
test_title_avg_w2v_vectors = []; # the avg-w2v for each sentence/review is stored in this list
for sentence in tqdm(x_test['clean_project_title']): # for each review/sentence
    vector = np.zeros(300) # as word vectors are of zero length
    cnt_words =0; # num of words with a valid vector in the sentence/review
    for word in sentence.split(): # for each word in a review/sentence
        if word in glove_words:
            vector += model[word]
            cnt_words += 1
    if cnt_words != 0:
        vector /= cnt_words
    test_title_avg_w2v_vectors.append(vector)

print(len(test_title_avg_w2v_vectors))
print(len(test_title_avg_w2v_vectors[0]))
100%|███████████████████████████████████████████████████████████████████████████| 9900/9900 [00:00<00:00, 38974.63it/s]
9900
300

1.5.2.3 Using Pretrained Models: TFIDF weighted W2V

In [62]:
# S = ["abc def pqr", "def def def abc", "pqr pqr def"]
tfidf_model = TfidfVectorizer()
tfidf_model.fit(x_train['clean_essay'])
# we are converting a dictionary with word as a key, and the idf as a value
dictionary = dict(zip(tfidf_model.get_feature_names(), list(tfidf_model.idf_)))
tfidf_words = set(tfidf_model.get_feature_names())
In [63]:
# average Word2Vec
# compute average word2vec for each review.
train_essay_tfidf_w2v_vectors = []; # the avg-w2v for each sentence/review is stored in this list
for sentence in tqdm(x_train['clean_essay']): # for each review/sentence
    vector = np.zeros(300) # as word vectors are of zero length
    tf_idf_weight =0; # num of words with a valid vector in the sentence/review
    for word in sentence.split(): # for each word in a review/sentence
        if (word in glove_words) and (word in tfidf_words):
            vec = model[word] # getting the vector for each word
            # here we are multiplying idf value(dictionary[word]) and the tf value((sentence.count(word)/len(sentence.split())))
            tf_idf = dictionary[word]*(sentence.count(word)/len(sentence.split())) # getting the tfidf value for each word
            vector += (vec * tf_idf) # calculating tfidf weighted w2v
            tf_idf_weight += tf_idf
    if tf_idf_weight != 0:
        vector /= tf_idf_weight
    train_essay_tfidf_w2v_vectors.append(vector)

print(len(train_essay_tfidf_w2v_vectors))
print(len(train_essay_tfidf_w2v_vectors[0]))
100%|███████████████████████████████████████████████████████████████████████████| 20100/20100 [01:02<00:00, 323.28it/s]
20100
300
In [64]:
# Similarly you can vectorize for title also
# average Word2Vec
# compute average word2vec for each review.
test_essay_tfidf_w2v_vectors = []; # the avg-w2v for each sentence/review is stored in this list
for sentence in tqdm(x_test['clean_essay']): # for each review/sentence
    vector = np.zeros(300) # as word vectors are of zero length
    tf_idf_weight =0; # num of words with a valid vector in the sentence/review
    for word in sentence.split(): # for each word in a review/sentence
        if (word in glove_words) and (word in tfidf_words):
            vec = model[word] # getting the vector for each word
            # here we are multiplying idf value(dictionary[word]) and the tf value((sentence.count(word)/len(sentence.split())))
            tf_idf = dictionary[word]*(sentence.count(word)/len(sentence.split())) # getting the tfidf value for each word
            vector += (vec * tf_idf) # calculating tfidf weighted w2v
            tf_idf_weight += tf_idf
    if tf_idf_weight != 0:
        vector /= tf_idf_weight
    test_essay_tfidf_w2v_vectors.append(vector)

print(len(test_essay_tfidf_w2v_vectors))
print(len(test_essay_tfidf_w2v_vectors[0]))
100%|█████████████████████████████████████████████████████████████████████████████| 9900/9900 [00:30<00:00, 319.64it/s]
9900
300

Proect_titles

In [65]:
# average Word2Vec
# compute average word2vec for each review.
train_title_tfidf_w2v_vectors = []; # the avg-w2v for each sentence/review is stored in this list
for sentence in tqdm(x_train['clean_project_title']): # for each review/sentence
    vector = np.zeros(300) # as word vectors are of zero length
    tf_idf_weight =0; # num of words with a valid vector in the sentence/review
    for word in sentence.split(): # for each word in a review/sentence
        if (word in glove_words) and (word in tfidf_words):
            vec = model[word] # getting the vector for each word
            # here we are multiplying idf value(dictionary[word]) and the tf value((sentence.count(word)/len(sentence.split())))
            tf_idf = dictionary[word]*(sentence.count(word)/len(sentence.split())) # getting the tfidf value for each word
            vector += (vec * tf_idf) # calculating tfidf weighted w2v
            tf_idf_weight += tf_idf
    if tf_idf_weight != 0:
        vector /= tf_idf_weight
    train_title_tfidf_w2v_vectors.append(vector)

print(len(train_title_tfidf_w2v_vectors))
print(len(train_title_tfidf_w2v_vectors[0]))
100%|█████████████████████████████████████████████████████████████████████████| 20100/20100 [00:00<00:00, 20849.39it/s]
20100
300
In [66]:
# average Word2Vec
# compute average word2vec for each review.
test_title_tfidf_w2v_vectors = []; # the avg-w2v for each sentence/review is stored in this list
for sentence in tqdm(x_test['clean_project_title']): # for each review/sentence
    vector = np.zeros(300) # as word vectors are of zero length
    tf_idf_weight =0; # num of words with a valid vector in the sentence/review
    for word in sentence.split(): # for each word in a review/sentence
        if (word in glove_words) and (word in tfidf_words):
            vec = model[word] # getting the vector for each word
            # here we are multiplying idf value(dictionary[word]) and the tf value((sentence.count(word)/len(sentence.split())))
            tf_idf = dictionary[word]*(sentence.count(word)/len(sentence.split())) # getting the tfidf value for each word
            vector += (vec * tf_idf) # calculating tfidf weighted w2v
            tf_idf_weight += tf_idf
    if tf_idf_weight != 0:
        vector /= tf_idf_weight
    test_title_tfidf_w2v_vectors.append(vector)

print(len(test_title_tfidf_w2v_vectors))
print(len(test_title_tfidf_w2v_vectors[0]))
100%|███████████████████████████████████████████████████████████████████████████| 9900/9900 [00:00<00:00, 20465.70it/s]
9900
300

1.5.3 Vectorizing Numerical features

In [67]:
price_data = resource_data.groupby('id').agg({'price':'sum', 'quantity':'sum'}).reset_index()
project_data = pd.merge(project_data, price_data, on='id', how='left')
print(price_data.head())
#print(project_data.columns)
print(x_train.columns)
        id    price  quantity
0  p000001   459.56         7
1  p000002   515.89        21
2  p000003   298.97         4
3  p000004  1113.69        98
4  p000005   485.99         8
Index(['Unnamed: 0', 'id', 'teacher_id', 'teacher_prefix', 'school_state',
       'project_submitted_datetime', 'project_grade_category',
       'project_essay_1', 'project_essay_2', 'project_essay_3',
       'project_essay_4', 'project_resource_summary',
       'teacher_number_of_previously_posted_projects', 'project_is_approved',
       'clean_categories', 'clean_subcategories', 'title_word_count',
       'essay_word_count', 'negitive', 'positive', 'neutral', 'compound',
       'clean_essay', 'clean_project_title'],
      dtype='object')
In [68]:
x_train = pd.merge(x_train, price_data, on = "id", how = "left")
x_test = pd.merge(x_test, price_data, on = "id", how = "left")

Standardize Price

In [69]:
# check this one: https://www.youtube.com/watch?v=0HOqOcln3Z4&t=530s
# standardization sklearn: https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.StandardScaler.html
from sklearn.preprocessing import StandardScaler

# price_standardized = standardScalar.fit(project_data['price'].values)
# this will rise the error
# ValueError: Expected 2D array, got 1D array instead: array=[725.05 213.03 329.   ... 399.   287.73   5.5 ].
# Reshape your data either using array.reshape(-1, 1)

price_scalar = StandardScaler()

price_scalar.fit(x_train['price'].values.reshape(-1,1)) # finding the mean and standard deviation of this data
print(f"TRAIN -> Mean : {price_scalar.mean_[0]}, Standard deviation : {np.sqrt(price_scalar.var_[0])}")
# Now standardize the data with above maen and variance.
train_price_standar = price_scalar.transform(x_train['price'].values.reshape(-1, 1))

price_scalar.fit(x_test['price'].values.reshape(-1,1)) # finding the mean and standard deviation of this data
print(f"TEST -> Mean : {price_scalar.mean_[0]}, Standard deviation : {np.sqrt(price_scalar.var_[0])}")
# Now standardize the data with above maen and variance.
test_price_standar = price_scalar.transform(x_test['price'].values.reshape(-1, 1))
TRAIN -> Mean : 297.7305975124378, Standard deviation : 367.7836935793401
TEST -> Mean : 300.35701919191916, Standard deviation : 392.20553854316364
In [70]:
print(train_price_standar.shape, y_train.shape)
print(test_price_standar.shape, y_test.shape)
(20100, 1) (20100,)
(9900, 1) (9900,)

Standardize Teacher previously posted Projects

In [71]:
warnings.filterwarnings("ignore")
price_scalar.fit(x_train['teacher_number_of_previously_posted_projects'].values.reshape(-1,1)) # finding the mean and standard deviation of this data
print(f"TRAIN -> Mean : {price_scalar.mean_[0]}, Standard deviation : {np.sqrt(price_scalar.var_[0])}")
# Now standardize the data with above maen and variance.
train_prev_proj_standar = price_scalar.transform(x_train['teacher_number_of_previously_posted_projects'].values.reshape(-1, 1))

price_scalar.fit(x_test['teacher_number_of_previously_posted_projects'].values.reshape(-1,1)) # finding the mean and standard deviation of this data
print(f"TEST -> Mean : {price_scalar.mean_[0]}, Standard deviation : {np.sqrt(price_scalar.var_[0])}")
# Now standardize the data with above maen and variance.
test_prev_proj_standar = price_scalar.transform(x_test['teacher_number_of_previously_posted_projects'].values.reshape(-1, 1))
TRAIN -> Mean : 11.233432835820896, Standard deviation : 27.845525139691485
TEST -> Mean : 11.167373737373737, Standard deviation : 28.099605773353765
In [72]:
print(train_prev_proj_standar.shape, y_train.shape)
print(test_prev_proj_standar.shape, y_test.shape)
(20100, 1) (20100,)
(9900, 1) (9900,)

Standardize Quantity

In [73]:
price_scalar.fit(x_train['quantity'].values.reshape(-1,1)) # finding the mean and standard deviation of this data
print(f"TRAIN -> Mean : {price_scalar.mean_[0]}, Standard deviation : {np.sqrt(price_scalar.var_[0])}")
# Now standardize the data with above maen and variance.
train_quantity_standar = price_scalar.transform(x_train['quantity'].values.reshape(-1, 1))

price_scalar.fit(x_test['quantity'].values.reshape(-1,1)) # finding the mean and standard deviation of this data
print(f"TEST -> Mean : {price_scalar.mean_[0]}, Standard deviation : {np.sqrt(price_scalar.var_[0])}")
# Now standardize the data with above maen and variance.
test_quantity_standar = price_scalar.transform(x_test['quantity'].values.reshape(-1, 1))
TRAIN -> Mean : 16.79268656716418, Standard deviation : 26.74389233749402
TEST -> Mean : 17.227474747474748, Standard deviation : 26.45436651603122
In [74]:
print(train_quantity_standar.shape, y_train.shape)
print(test_quantity_standar.shape, y_test.shape)
(20100, 1) (20100,)
(9900, 1) (9900,)

Standardize Title_word_count

In [75]:
title_scalar = StandardScaler()
title_scalar.fit(x_train['title_word_count'].values.reshape(-1,1)) # finding the mean and standard deviation of this data
print(f"Mean : {title_scalar.mean_[0]}, Standard deviation : {np.sqrt(price_scalar.var_[0])}")
train_title_word_count_standar = title_scalar.transform(x_train['title_word_count'].values.reshape(-1, 1))

title_scalar.fit(x_test['title_word_count'].values.reshape(-1,1)) # finding the mean and standard deviation of this data
print(f"Mean : {title_scalar.mean_[0]}, Standard deviation : {np.sqrt(price_scalar.var_[0])}")
test_title_word_count_standar = title_scalar.transform(x_test['title_word_count'].values.reshape(-1, 1))

print(train_title_word_count_standar.shape, y_train.shape)
print(test_title_word_count_standar.shape, y_test.shape)
Mean : 5.220149253731344, Standard deviation : 26.45436651603122
Mean : 5.21030303030303, Standard deviation : 26.45436651603122
(20100, 1) (20100,)
(9900, 1) (9900,)

Standardize Essay_word_count

In [76]:
essay_scalar = StandardScaler()

essay_scalar.fit(x_train['essay_word_count'].values.reshape(-1,1)) # finding the mean and standard deviation of this data
train_essay_word_count_standar = essay_scalar.transform(x_train['essay_word_count'].values.reshape(-1, 1))

essay_scalar.fit(x_train['essay_word_count'].values.reshape(-1,1)) # finding the mean and standard deviation of this data
test_essay_word_count_standar = essay_scalar.transform(x_test['essay_word_count'].values.reshape(-1, 1))

print(train_essay_word_count_standar.shape, y_train.shape)
print(test_essay_word_count_standar.shape, y_test.shape)
(20100, 1) (20100,)
(9900, 1) (9900,)

Standardize Positive Intensity

In [77]:
essay_scalar.fit(x_train['positive'].values.reshape(-1,1)) # finding the mean and standard deviation of this data
train_positive_standar = essay_scalar.transform(x_train['positive'].values.reshape(-1, 1))

essay_scalar.fit(x_train['positive'].values.reshape(-1,1)) # finding the mean and standard deviation of this data
test_positive_standar = essay_scalar.transform(x_test['positive'].values.reshape(-1, 1))

print(train_positive_standar.shape, y_train.shape)
print(test_positive_standar.shape, y_test.shape)
(20100, 1) (20100,)
(9900, 1) (9900,)

Standarsize Negitive Intensity

In [78]:
essay_scalar.fit(x_train['negitive'].values.reshape(-1,1)) # finding the mean and standard deviation of this data
train_negitive_standar = essay_scalar.transform(x_train['negitive'].values.reshape(-1, 1))

essay_scalar.fit(x_train['negitive'].values.reshape(-1,1)) # finding the mean and standard deviation of this data
test_negitive_standar = essay_scalar.transform(x_test['negitive'].values.reshape(-1, 1))

print(train_negitive_standar.shape, y_train.shape)
print(test_negitive_standar.shape, y_test.shape)
(20100, 1) (20100,)
(9900, 1) (9900,)

Standardize Neutral Intensity

In [79]:
essay_scalar.fit(x_train['neutral'].values.reshape(-1,1)) # finding the mean and standard deviation of this data
train_neutral_standar = essay_scalar.transform(x_train['neutral'].values.reshape(-1, 1))

essay_scalar.fit(x_train['neutral'].values.reshape(-1,1))
test_neutral_standar = essay_scalar.transform(x_test['neutral'].values.reshape(-1, 1))

print(train_neutral_standar.shape, y_train.shape)
print(test_neutral_standar.shape, y_test.shape)
(20100, 1) (20100,)
(9900, 1) (9900,)

Assignment 9: RF & GBDT

The response tabel is built only on train dataset. For a category which is not there in train data and present in test data, we will encode them with default values Ex: in our test data if have State: D then we encode it as [0.5, 0.05]

  1. Apply both Random Forrest and GBDT on these feature sets
    • Set 1: categorical(instead of one hot encoding, try response coding: use probability values), numerical features + project_title(BOW) + preprocessed_eassay (BOW)
    • Set 2: categorical(instead of one hot encoding, try response coding: use probability values), numerical features + project_title(TFIDF)+ preprocessed_eassay (TFIDF)
    • Set 3: categorical(instead of one hot encoding, try response coding: use probability values), numerical features + project_title(AVG W2V)+ preprocessed_eassay (AVG W2V)
    • Set 4: categorical(instead of one hot encoding, try response coding: use probability values), numerical features + project_title(TFIDF W2V)+ preprocessed_eassay (TFIDF W2V)

  2. The hyper paramter tuning (Consider any two hyper parameters preferably n_estimators, max_depth)
    • Find the best hyper parameter which will give the maximum AUC value
    • find the best hyper paramter using k-fold cross validation/simple cross validation data
    • use gridsearch cv or randomsearch cv or you can write your own for loops to do this task

  3. Representation of results
    • You need to plot the performance of model both on train data and cross validation data for each hyper parameter, like shown in the figure with X-axis as n_estimators, Y-axis as max_depth, and Z-axis as AUC Score , we have given the notebook which explains how to plot this 3d plot, you can find it in the same drive 3d_scatter_plot.ipynb
    • or


    • You need to plot the performance of model both on train data and cross validation data for each hyper parameter, like shown in the figure seaborn heat maps with rows as n_estimators, columns as max_depth, and values inside the cell representing AUC Score
    • You can choose either of the plotting techniques: 3d plot or heat map
    • Once after you found the best hyper parameter, you need to train your model with it, and find the AUC on test data and plot the ROC curve on both train and test.
    • Along with plotting ROC curve, you need to print the confusion matrix with predicted and original labels of test data points

  4. Conclusion
    • You need to summarize the results at the end of the notebook, summarize it in the table format. To print out a table please refer to this prettytable library link

Note: Data Leakage

  1. There will be an issue of data-leakage if you vectorize the entire data and then split it into train/cv/test.
  2. To avoid the issue of data-leakage, make sure to split your data first and then vectorize it.
  3. While vectorizing your data, apply the method fit_transform() on you train data, and apply the method transform() on cv/test data.
  4. For more details please go through this link.

Random Forest and GBDT

2.1 RF and GBDT Set-1

Merging all Categorical and Numerical _ SET-1 BOW Encoding

In [80]:
print(X_train_clean_cat_ohe.shape)
print(X_train_clean_subcat_ohe.shape)
print(X_train_grade_ohe.shape)
print(X_train_state_ohe.shape)
print(X_train_teacher_ohe.shape)
print(train_text_bow.shape)
print(test_text_bow.shape)
(20100, 2)
(20100, 2)
(20100, 2)
(20100, 2)
(20100, 2)
(20100, 8528)
(9900, 8528)
In [81]:
from scipy.sparse import hstack
# with the same hstack function we are concatinating a sparse matrix and a dense matirx :)
X_train1 = hstack((X_train_clean_cat_ohe, X_train_clean_subcat_ohe, X_train_grade_ohe, X_train_state_ohe, X_train_teacher_ohe, train_text_bow
                   ,train_title_bow)).tocsr()

X_test1 = hstack((X_test_clean_cat_ohe, X_test_clean_subcat_ohe, X_test_grade_ohe, X_test_state_ohe, X_test_teacher_ohe, test_text_bow
                   ,test_title_bow)).tocsr()

print(X_train1.shape, y_train.shape)
print(X_test1.shape, y_test.shape)
print(type(X_train1))
# train_text_tfidf, train_title_tfidf, train_avg_w2v_vectors, train_title_avg_w2v_vectors,train_essay_tfidf_w2v_vectors, train_title_tfidf_w2v_vectors
#test_text_tfidf, test_title_tfidf, test_avg_w2v_vectors, test_title_avg_w2v_vectors,test_essay_tfidf_w2v_vectors, test_title_tfidf_w2v_vectors
(20100, 9691) (20100,)
(9900, 9691) (9900,)
<class 'scipy.sparse.csr.csr_matrix'>
In [82]:
print(X_train1.shape, y_train.shape)
(20100, 9691) (20100,)

Hyperparameter Tunning GridSearch

In [90]:
from sklearn.model_selection import GridSearchCV
from sklearn.ensemble import RandomForestClassifier
import seaborn as sea
In [91]:
RF = RandomForestClassifier()

parameters = {'max_depth': [1, 5, 10, 50, 100, 500], 'min_samples_split': [5, 10, 100, 500]}

classifier = GridSearchCV(RF, parameters, cv=3, scoring='roc_auc')
classifier.fit(X_train1, y_train)
Out[91]:
GridSearchCV(cv=3, error_score='raise-deprecating',
       estimator=RandomForestClassifier(bootstrap=True, class_weight=None, criterion='gini',
            max_depth=None, max_features='auto', max_leaf_nodes=None,
            min_impurity_decrease=0.0, min_impurity_split=None,
            min_samples_leaf=1, min_samples_split=2,
            min_weight_fraction_leaf=0.0, n_estimators='warn', n_jobs=None,
            oob_score=False, random_state=None, verbose=0,
            warm_start=False),
       fit_params=None, iid='warn', n_jobs=None,
       param_grid={'max_depth': [1, 5, 10, 50, 100, 500], 'min_samples_split': [5, 10, 100, 500]},
       pre_dispatch='2*n_jobs', refit=True, return_train_score='warn',
       scoring='roc_auc', verbose=0)
In [92]:
max_scores = pd.DataFrame(classifier.cv_results_).groupby(['param_min_samples_split', 'param_max_depth']).max().unstack()[['mean_test_score', 'mean_train_score']]

fig, ax = plt.subplots(1,2, figsize=(20,6))

sea.heatmap(max_scores.mean_train_score, annot = True, fmt='.4g', ax=ax[0])
sea.heatmap(max_scores.mean_test_score, annot = True, fmt='.4g', ax=ax[1])

ax[0].set_title('Train Set')
ax[1].set_title('CV Set')

plt.show()

From depth 100 and sample split 100, we are gtting accuracy greater than 90

In [93]:
#################################PLOTly Plot##############################
#x1 = [1, 5, 10, 50, 100, 500]
#y1 = [5, 10, 100, 500]
#z1 = [0.5475, 0.6478, 0.702, 0.9442]
# https://plot.ly/python/3d-axes/
#trace1 = go.Scatter3d(x=x1,y=y1,z=z1, name = 'train')
#trace2 = go.Scatter3d(x=x2,y=y2,z=z2, name = 'Cross validation')
#data = [trace1]

#layout = go.Layout(scene = dict(
#        xaxis = dict(title='max_depth'),
#        yaxis = dict(title='min_samples_split'),
#        zaxis = dict(title='AUC'),))
#
#fig = go.Figure(data=data, layout=layout)
#offline.iplot(fig, filename='3d-scatter-colorscale')

Trainng Model Using Best HyperParameter

In [94]:
def batch_predict(clf, data):
    # roc_auc_score(y_true, y_score) the 2nd parameter should be probability estimates of the positive class
    # not the predicted outputs
    
    y_data_pred = []
    tr_loop = data.shape[0] - data.shape[0]%1000
    # consider you X_tr shape is 49041, then your cr_loop will be 49041 - 49041%1000 = 49000
    # in this for loop we will iterate unti the last 1000 multiplier
    for i in range(0, tr_loop, 1000):
        y_data_pred.extend(clf.predict_proba(data[i:i+1000])[:,1])
    # we will be predicting for the last data points
    y_data_pred.extend(clf.predict_proba(data[tr_loop:])[:,1])
    
    return y_data_pred
In [121]:
# https://scikit-learn.org/stable/modules/generated/sklearn.metrics.roc_curve.html#sklearn.metrics.roc_curve
from sklearn.metrics import roc_curve, auc
from sklearn.tree import DecisionTreeClassifier

classifier = DecisionTreeClassifier(max_depth = 10, min_samples_split = 250)

classifier.fit(X_train1, y_train)
#classifier_viz.fit(X_train1, y_train)

y_train_pred = batch_predict(classifier,X_train1)  
y_test_pred = batch_predict(classifier,X_test1)
#y_test_pred = batch_predict(classifier_viz,X_test1)

train_fpr, train_tpr, tr_thresholds = roc_curve(y_train, y_train_pred)
test_fpr, test_tpr, te_thresholds = roc_curve(y_test, y_test_pred)

plt.plot(train_fpr, train_tpr, label="Train AUC ="+str(auc(train_fpr, train_tpr)))
plt.plot(test_fpr, test_tpr, label="Test AUC ="+str(auc(test_fpr, test_tpr)))
plt.legend()
plt.xlabel("True Positive Rate(TPR)")
plt.ylabel("False Positive Rate(FPR)")
plt.title("AUC")
plt.grid()
plt.show()

Confusion Matrix for Train and Test Data

In [97]:
def predict(proba, threshould, fpr, tpr):
    
    t = threshould[np.argmax(fpr*(1-tpr))]
    print("the maximum value of tpr*(1-fpr)", max(tpr*(1-fpr)), "for threshold", np.round(t,3))
    predictions = []
    global predictions1
    for i in proba:
        if i>=t:
            predictions.append(1)
        else:
            predictions.append(0)
    predictions1 = predictions
    return predictions
In [98]:
#https://www.quantinsti.com/blog/creating-heatmap-using-python-seaborn
import seaborn as sns; sns.set()

con_m_train = confusion_matrix(y_train, predict(y_train_pred, tr_thresholds, train_fpr, train_tpr))
con_m_test = confusion_matrix(y_test, predict(y_test_pred, te_thresholds, test_fpr, test_tpr))

key = (np.asarray([['TN','FP'], ['FN', 'TP']]))
fig, ax = plt.subplots(1,2, figsize=(12,5)) 

labels_train = (np.asarray(["{0} = {1:.2f}" .format(key, value) for key, value in zip(key.flatten(), con_m_train.flatten())])).reshape(2,2)
labels_test = (np.asarray(["{0} = {1:.2f}" .format(key, value) for key, value in zip(key.flatten(), con_m_test.flatten())])).reshape(2,2)

sns.heatmap(con_m_train, linewidths=.5, xticklabels=['PREDICTED : NO', 'PREDICTED : YES'], yticklabels=['ACTUAL : NO', 'ACTUAL : YES'], annot = labels_train, fmt = '', ax=ax[0])
sns.heatmap(con_m_test, linewidths=.5, xticklabels=['PREDICTED : NO', 'PREDICTED : YES'], yticklabels=['ACTUAL : NO', 'ACTUAL : YES'], annot = labels_test, fmt = '', ax=ax[1])

ax[0].set_title('Train Set')
ax[1].set_title('Test Set')

plt.show()
the maximum value of tpr*(1-fpr) 0.4788768424287167 for threshold 0.902
the maximum value of tpr*(1-fpr) 0.36361669095437693 for threshold 0.914

2.2 Merging all Categorical and Numerical _ SET-2 TF-IDF Encoding

In [83]:
from scipy.sparse import hstack
# with the same hstack function we are concatinating a sparse matrix and a dense matirx :)
X_train2 = hstack((X_train_clean_cat_ohe, X_train_clean_subcat_ohe, X_train_grade_ohe, X_train_state_ohe, X_train_teacher_ohe,
                   train_text_tfidf, train_title_tfidf)).tocsr()

X_test2 = hstack((X_test_clean_cat_ohe, X_test_clean_subcat_ohe, X_test_grade_ohe, X_test_state_ohe, X_test_teacher_ohe,
                  test_text_tfidf, test_title_tfidf)).tocsr()

print(X_train2.shape, y_train.shape)
print(X_test2.shape, y_test.shape)
print(type(X_train2))
# train_text_tfidf, train_title_tfidf, train_avg_w2v_vectors, train_title_avg_w2v_vectors,train_essay_tfidf_w2v_vectors, train_title_tfidf_w2v_vectors
#test_text_tfidf, test_title_tfidf, test_avg_w2v_vectors, test_title_avg_w2v_vectors,test_essay_tfidf_w2v_vectors, test_title_tfidf_w2v_vectors
(20100, 9691) (20100,)
(9900, 9691) (9900,)
<class 'scipy.sparse.csr.csr_matrix'>

Hyperparameter Tunning

In [84]:
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import GridSearchCV
RF = RandomForestClassifier()

parameters = {'max_depth': [1, 5, 10, 50, 100, 500], 'min_samples_split': [5, 10, 100, 500]}

classifier = GridSearchCV(RF, parameters, cv=3, scoring='roc_auc')
classifier.fit(X_train2, y_train)
Out[84]:
GridSearchCV(cv=3, error_score='raise-deprecating',
       estimator=RandomForestClassifier(bootstrap=True, class_weight=None, criterion='gini',
            max_depth=None, max_features='auto', max_leaf_nodes=None,
            min_impurity_decrease=0.0, min_impurity_split=None,
            min_samples_leaf=1, min_samples_split=2,
            min_weight_fraction_leaf=0.0, n_estimators='warn', n_jobs=None,
            oob_score=False, random_state=None, verbose=0,
            warm_start=False),
       fit_params=None, iid='warn', n_jobs=None,
       param_grid={'max_depth': [1, 5, 10, 50, 100, 500], 'min_samples_split': [5, 10, 100, 500]},
       pre_dispatch='2*n_jobs', refit=True, return_train_score='warn',
       scoring='roc_auc', verbose=0)
In [85]:
import seaborn as sea
max_scores = pd.DataFrame(classifier.cv_results_).groupby(['param_min_samples_split', 'param_max_depth']).max().unstack()[['mean_test_score', 'mean_train_score']]

fig, ax = plt.subplots(1,2, figsize=(20,6))
sea.heatmap(max_scores.mean_train_score, annot = True, fmt='.4g', ax=ax[0])
sea.heatmap(max_scores.mean_test_score, annot = True, fmt='.4g', ax=ax[1])

ax[0].set_title('Train Set')
ax[1].set_title('CV Set')

plt.show()

Training Model using best Hyperparameter

In [112]:
# https://scikit-learn.org/stable/modules/generated/sklearn.metrics.roc_curve.html#sklearn.metrics.roc_curve
from sklearn.metrics import roc_curve, auc
from sklearn.tree import DecisionTreeClassifier

classifier = DecisionTreeClassifier(max_depth = 10, min_samples_split = 250)

classifier.fit(X_train2, y_train)
#classifier_viz.fit(X_train1, y_train)

y_train_pred = batch_predict(classifier,X_train2)  
y_test_pred = batch_predict(classifier,X_test2)
#y_test_pred = batch_predict(classifier_viz,X_test1)

train_fpr, train_tpr, tr_thresholds = roc_curve(y_train, y_train_pred)
test_fpr, test_tpr, te_thresholds = roc_curve(y_test, y_test_pred)

plt.plot(train_fpr, train_tpr, label="Train AUC ="+str(auc(train_fpr, train_tpr)))
plt.plot(test_fpr, test_tpr, label="Test AUC ="+str(auc(test_fpr, test_tpr)))
plt.legend()
plt.xlabel("True Positive Rate(TPR)")
plt.ylabel("False Positive Rate(FPR)")
plt.title("AUC")
plt.grid()
plt.show()

Confusion Matrix For Train & Test Data

In [101]:
#https://www.quantinsti.com/blog/creating-heatmap-using-python-seaborn
import seaborn as sns; sns.set()

con_m_train = confusion_matrix(y_train, predict(y_train_pred, tr_thresholds, train_fpr, train_tpr))
con_m_test = confusion_matrix(y_test, predict(y_test_pred, te_thresholds, test_fpr, test_tpr))

key = (np.asarray([['TN','FP'], ['FN', 'TP']]))
fig, ax = plt.subplots(1,2, figsize=(12,5)) 

labels_train = (np.asarray(["{0} = {1:.2f}" .format(key, value) for key, value in zip(key.flatten(), con_m_train.flatten())])).reshape(2,2)
labels_test = (np.asarray(["{0} = {1:.2f}" .format(key, value) for key, value in zip(key.flatten(), con_m_test.flatten())])).reshape(2,2)

sns.heatmap(con_m_train, linewidths=.5, xticklabels=['PREDICTED : NO', 'PREDICTED : YES'], yticklabels=['ACTUAL : NO', 'ACTUAL : YES'], annot = labels_train, fmt = '', ax=ax[0])
sns.heatmap(con_m_test, linewidths=.5, xticklabels=['PREDICTED : NO', 'PREDICTED : YES'], yticklabels=['ACTUAL : NO', 'ACTUAL : YES'], annot = labels_test, fmt = '', ax=ax[1])

ax[0].set_title('Train Set')
ax[1].set_title('Test Set')

plt.show()
the maximum value of tpr*(1-fpr) 0.5174641982787899 for threshold 0.921
the maximum value of tpr*(1-fpr) 0.34512947997353216 for threshold 0.92

2.3 Merging all Categorical and Numerical _ SET-3 AVG-W2V Encoding

In [84]:
X_train3 = hstack((X_train_clean_cat_ohe, X_train_clean_subcat_ohe, X_train_grade_ohe, X_train_state_ohe, X_train_teacher_ohe,
                   train_avg_w2v_vectors, train_title_avg_w2v_vectors)).tocsr()

X_test3 = hstack((X_test_clean_cat_ohe, X_test_clean_subcat_ohe, X_test_grade_ohe, X_test_state_ohe, X_test_teacher_ohe,
                  test_avg_w2v_vectors, test_title_avg_w2v_vectors)).tocsr()

print(X_train3.shape, y_train.shape)
print(X_test3.shape, y_test.shape)
print(type(X_train3))
# train_text_tfidf, train_title_tfidf, train_avg_w2v_vectors, train_title_avg_w2v_vectors,train_essay_tfidf_w2v_vectors, train_title_tfidf_w2v_vectors
#test_text_tfidf, test_title_tfidf, test_avg_w2v_vectors, test_title_avg_w2v_vectors,test_essay_tfidf_w2v_vectors, test_title_tfidf_w2v_vectors
(20100, 610) (20100,)
(9900, 610) (9900,)
<class 'scipy.sparse.csr.csr_matrix'>
In [85]:
print(X_train3.shape, y_train.shape)
(20100, 610) (20100,)

Hyperparameter Tunning

In [119]:
RF = RandomForestClassifier()

parameters = {'max_depth': [1, 5, 10, 50, 100, 500], 'min_samples_split': [5, 10, 100, 500]}

classifier = GridSearchCV(RF, parameters, cv=3, scoring='roc_auc')
classifier.fit(X_train3, y_train)
Out[119]:
GridSearchCV(cv=3, error_score='raise-deprecating',
       estimator=RandomForestClassifier(bootstrap=True, class_weight=None, criterion='gini',
            max_depth=None, max_features='auto', max_leaf_nodes=None,
            min_impurity_decrease=0.0, min_impurity_split=None,
            min_samples_leaf=1, min_samples_split=2,
            min_weight_fraction_leaf=0.0, n_estimators='warn', n_jobs=None,
            oob_score=False, random_state=None, verbose=0,
            warm_start=False),
       fit_params=None, iid='warn', n_jobs=None,
       param_grid={'max_depth': [1, 5, 10, 50, 100, 500], 'min_samples_split': [5, 10, 100, 500]},
       pre_dispatch='2*n_jobs', refit=True, return_train_score='warn',
       scoring='roc_auc', verbose=0)
In [120]:
max_scores = pd.DataFrame(classifier.cv_results_).groupby(['param_min_samples_split', 'param_max_depth']).max().unstack()[['mean_test_score', 'mean_train_score']]

fig, ax = plt.subplots(1,2, figsize=(20,6))

sea.heatmap(max_scores.mean_train_score, annot = True, fmt='.4g', ax=ax[0])
sea.heatmap(max_scores.mean_test_score, annot = True, fmt='.4g', ax=ax[1])

ax[0].set_title('Train Set')
ax[1].set_title('CV Set')

plt.show()

Training Model Using Hyperparameter

In [113]:
# https://scikit-learn.org/stable/modules/generated/sklearn.metrics.roc_curve.html#sklearn.metrics.roc_curve
from sklearn.metrics import roc_curve, auc
from sklearn.tree import DecisionTreeClassifier

classifier = DecisionTreeClassifier(max_depth = 5, min_samples_split = 500)

classifier.fit(X_train3, y_train)
#classifier_viz.fit(X_train1, y_train)

y_train_pred = batch_predict(classifier,X_train3)  
y_test_pred = batch_predict(classifier,X_test3)
#y_test_pred = batch_predict(classifier_viz,X_test1)

train_fpr, train_tpr, tr_thresholds = roc_curve(y_train, y_train_pred)
test_fpr, test_tpr, te_thresholds = roc_curve(y_test, y_test_pred)

plt.plot(train_fpr, train_tpr, label="Train AUC ="+str(auc(train_fpr, train_tpr)))
plt.plot(test_fpr, test_tpr, label="Test AUC ="+str(auc(test_fpr, test_tpr)))
plt.legend()
plt.xlabel("True Positive Rate(TPR)")
plt.ylabel("False Positive Rate(FPR)")
plt.title("AUC")
plt.grid()
plt.show()

Train & Test Confusion Matrix

In [114]:
#https://www.quantinsti.com/blog/creating-heatmap-using-python-seaborn
import seaborn as sns
sns.set()

con_m_train = confusion_matrix(y_train, predict(y_train_pred, tr_thresholds, train_fpr, train_tpr))
con_m_test = confusion_matrix(y_test, predict(y_test_pred, te_thresholds, test_fpr, test_tpr))

key = (np.asarray([['TN','FP'], ['FN', 'TP']]))
fig, ax = plt.subplots(1,2, figsize=(12,5)) 

labels_train = (np.asarray(["{0} = {1:.2f}" .format(key, value) for key, value in zip(key.flatten(), con_m_train.flatten())])).reshape(2,2)
labels_test = (np.asarray(["{0} = {1:.2f}" .format(key, value) for key, value in zip(key.flatten(), con_m_test.flatten())])).reshape(2,2)

sns.heatmap(con_m_train, linewidths=.5, xticklabels=['PREDICTED : NO', 'PREDICTED : YES'], yticklabels=['ACTUAL : NO', 'ACTUAL : YES'], annot = labels_train, fmt = '', ax=ax[0])
sns.heatmap(con_m_test, linewidths=.5, xticklabels=['PREDICTED : NO', 'PREDICTED : YES'], yticklabels=['ACTUAL : NO', 'ACTUAL : YES'], annot = labels_test, fmt = '', ax=ax[1])

ax[0].set_title('Train Set')
ax[1].set_title('Test Set')

plt.show()
the maximum value of tpr*(1-fpr) 0.35700334632392833 for threshold 0.855
the maximum value of tpr*(1-fpr) 0.32063718281625825 for threshold 0.855

2.4 Merging all Categorical and Numerical _ SET-4 TFIDF-W2V Encoding

In [86]:
from scipy.sparse import hstack
# with the same hstack function we are concatinating a sparse matrix and a dense matirx :)
X_train4 = hstack((X_train_clean_cat_ohe, X_train_clean_subcat_ohe, X_train_grade_ohe, X_train_state_ohe, X_train_teacher_ohe,
                   train_essay_tfidf_w2v_vectors, train_title_tfidf_w2v_vectors)).tocsr()

X_test4 = hstack((X_test_clean_cat_ohe, X_test_clean_subcat_ohe, X_test_grade_ohe, X_test_state_ohe, X_test_teacher_ohe,
                  test_essay_tfidf_w2v_vectors, test_title_tfidf_w2v_vectors)).tocsr()

print(X_train4.shape, y_train.shape)
print(X_test4.shape, y_test.shape)
print(type(X_train2))
# train_text_tfidf, train_title_tfidf, train_avg_w2v_vectors, train_title_avg_w2v_vectors,train_essay_tfidf_w2v_vectors, train_title_tfidf_w2v_vectors
#test_text_tfidf, test_title_tfidf, test_avg_w2v_vectors, test_title_avg_w2v_vectors,test_essay_tfidf_w2v_vectors, test_title_tfidf_w2v_vectors
(20100, 610) (20100,)
(9900, 610) (9900,)
<class 'scipy.sparse.csr.csr_matrix'>
In [87]:
print(X_train4.shape, y_train.shape)
(20100, 610) (20100,)

Hyperparameter Tunning

In [128]:
RF = RandomForestClassifier()

parameters = {'max_depth': [1, 5, 10, 50, 100, 500], 'min_samples_split': [5, 10, 100, 500, 1000, 1250]}

classifier = GridSearchCV(RF, parameters, cv=3, scoring='roc_auc')
classifier.fit(X_train4, y_train)
Out[128]:
GridSearchCV(cv=3, error_score='raise-deprecating',
       estimator=RandomForestClassifier(bootstrap=True, class_weight=None, criterion='gini',
            max_depth=None, max_features='auto', max_leaf_nodes=None,
            min_impurity_decrease=0.0, min_impurity_split=None,
            min_samples_leaf=1, min_samples_split=2,
            min_weight_fraction_leaf=0.0, n_estimators='warn', n_jobs=None,
            oob_score=False, random_state=None, verbose=0,
            warm_start=False),
       fit_params=None, iid='warn', n_jobs=None,
       param_grid={'max_depth': [1, 5, 10, 50, 100, 500], 'min_samples_split': [5, 10, 100, 500, 1000, 1250]},
       pre_dispatch='2*n_jobs', refit=True, return_train_score='warn',
       scoring='roc_auc', verbose=0)
In [129]:
max_scores = pd.DataFrame(classifier.cv_results_).groupby(['param_min_samples_split', 'param_max_depth']).max().unstack()[['mean_test_score', 'mean_train_score']]

fig, ax = plt.subplots(1,2, figsize=(20,6))

sea.heatmap(max_scores.mean_train_score, annot = True, fmt='.4g', ax=ax[0])
sea.heatmap(max_scores.mean_test_score, annot = True, fmt='.4g', ax=ax[1])

ax[0].set_title('Train Set')
ax[1].set_title('CV Set')

plt.show()

Training Model using Best Hyperparameter

In [118]:
# https://scikit-learn.org/stable/modules/generated/sklearn.metrics.roc_curve.html#sklearn.metrics.roc_curve
from sklearn.metrics import roc_curve, auc
from sklearn.tree import DecisionTreeClassifier

classifier = DecisionTreeClassifier(max_depth = 5, min_samples_split = 500)

classifier.fit(X_train4, y_train)
#classifier_viz.fit(X_train1, y_train)

y_train_pred = batch_predict(classifier,X_train4)  
y_test_pred = batch_predict(classifier,X_test4)
#y_test_pred = batch_predict(classifier_viz,X_test1)

train_fpr, train_tpr, tr_thresholds = roc_curve(y_train, y_train_pred)
test_fpr, test_tpr, te_thresholds = roc_curve(y_test, y_test_pred)

plt.plot(train_fpr, train_tpr, label="Train AUC ="+str(auc(train_fpr, train_tpr)))
plt.plot(test_fpr, test_tpr, label="Test AUC ="+str(auc(test_fpr, test_tpr)))
plt.legend()
plt.xlabel("True Positive Rate(TPR)")
plt.ylabel("False Positive Rate(FPR)")
plt.title("AUC")
plt.grid()
plt.show()

Train & Test Confusion Matrix

In [119]:
#https://www.quantinsti.com/blog/creating-heatmap-using-python-seaborn
import seaborn as sns
sns.set()

con_m_train = confusion_matrix(y_train, predict(y_train_pred, tr_thresholds, train_fpr, train_tpr))
con_m_test = confusion_matrix(y_test, predict(y_test_pred, te_thresholds, test_fpr, test_tpr))

key = (np.asarray([['TN','FP'], ['FN', 'TP']]))
fig, ax = plt.subplots(1,2, figsize=(12,5)) 

labels_train = (np.asarray(["{0} = {1:.2f}" .format(key, value) for key, value in zip(key.flatten(), con_m_train.flatten())])).reshape(2,2)
labels_test = (np.asarray(["{0} = {1:.2f}" .format(key, value) for key, value in zip(key.flatten(), con_m_test.flatten())])).reshape(2,2)

sns.heatmap(con_m_train, linewidths=.5, xticklabels=['PREDICTED : NO', 'PREDICTED : YES'], yticklabels=['ACTUAL : NO', 'ACTUAL : YES'], annot = labels_train, fmt = '', ax=ax[0])
sns.heatmap(con_m_test, linewidths=.5, xticklabels=['PREDICTED : NO', 'PREDICTED : YES'], yticklabels=['ACTUAL : NO', 'ACTUAL : YES'], annot = labels_test, fmt = '', ax=ax[1])

ax[0].set_title('Train Set')
ax[1].set_title('Test Set')

plt.show()
the maximum value of tpr*(1-fpr) 0.37570050646710457 for threshold 0.879
the maximum value of tpr*(1-fpr) 0.34062407499627606 for threshold 0.851

2.5 Apply Gradient Boosted Decision Trees (GBDT)

Applying Gradient Boosted Decision Tree on same sets again to verify which algo is performing better, we need to do hyperparameter tuning corresponding to the model you selected and procedure in step 2 and step 3 </br>

2.5.1 Applying GBDT on Set - 1

In [88]:
print(X_train1.shape, y_train.shape)
(20100, 9691) (20100,)

Hyperparameter Tunning using GridSearchCV

In [96]:
from sklearn.model_selection import GridSearchCV
from sklearn.ensemble import GradientBoostingClassifier
In [139]:
GBDT = GradientBoostingClassifier()

parameters = {'max_depth' : [5, 10, 20, 50, 100], 'n_estimators' : [10, 25, 50, 100, 250]}

classifier = GridSearchCV(GBDT, parameters, cv=3, scoring = 'roc_auc')
classifier.fit(X_train1, y_train)
Out[139]:
GridSearchCV(cv=3, error_score='raise-deprecating',
       estimator=GradientBoostingClassifier(criterion='friedman_mse', init=None,
              learning_rate=0.1, loss='deviance', max_depth=3,
              max_features=None, max_leaf_nodes=None,
              min_impurity_decrease=0.0, min_impurity_split=None,
              min_samples_leaf=1, min_sampl...      subsample=1.0, tol=0.0001, validation_fraction=0.1,
              verbose=0, warm_start=False),
       fit_params=None, iid='warn', n_jobs=None,
       param_grid={'max_depth': [5, 10, 20, 50, 100], 'n_estimators': [10, 25, 50, 100, 250]},
       pre_dispatch='2*n_jobs', refit=True, return_train_score='warn',
       scoring='roc_auc', verbose=0)
In [143]:
max_scores = pd.DataFrame(classifier.cv_results_).groupby(['param_n_estimators', 'param_max_depth']).max().unstack()[['mean_test_score', 'mean_train_score']]

fig, ax = plt.subplots(1,2, figsize=(20,6))

sea.heatmap(max_scores.mean_train_score, annot = True, fmt='.4g', ax=ax[0])
sea.heatmap(max_scores.mean_test_score, annot = True, fmt='.4g', ax=ax[1])

ax[0].set_title('Train Set')
ax[1].set_title('CV Set')

plt.show()

Training Using Best Hyperparameter

In [123]:
# https://scikit-learn.org/stable/modules/generated/sklearn.metrics.roc_curve.html#sklearn.metrics.roc_curve
from sklearn.metrics import roc_curve, auc
from sklearn.ensemble import GradientBoostingClassifier

classifier = GradientBoostingClassifier(max_depth = 5 , min_samples_split = 250)

classifier.fit(X_train1, y_train)
#classifier_viz.fit(X_train1, y_train)

y_train_pred = batch_predict(classifier,X_train1)
y_test_pred = batch_predict(classifier,X_test1)
#y_test_pred = batch_predict(classifier_viz,X_test1)

train_fpr, train_tpr, tr_thresholds = roc_curve(y_train, y_train_pred)
test_fpr, test_tpr, te_thresholds = roc_curve(y_test, y_test_pred)

plt.plot(train_fpr, train_tpr, label="Train AUC ="+str(auc(train_fpr, train_tpr)))
plt.plot(test_fpr, test_tpr, label="Test AUC ="+str(auc(test_fpr, test_tpr)))
plt.legend()
plt.xlabel("True Positive Rate(TPR)")
plt.ylabel("False Positive Rate(FPR)")
plt.title("AUC")
plt.grid()
plt.show()

Confusion Matrix

In [124]:
#https://www.quantinsti.com/blog/creating-heatmap-using-python-seaborn
import seaborn as sns
sns.set()

con_m_train = confusion_matrix(y_train, predict(y_train_pred, tr_thresholds, train_fpr, train_tpr))
con_m_test = confusion_matrix(y_test, predict(y_test_pred, te_thresholds, test_fpr, test_tpr))

key = (np.asarray([['TN','FP'], ['FN', 'TP']]))
fig, ax = plt.subplots(1,2, figsize=(12,5)) 

labels_train = (np.asarray(["{0} = {1:.2f}" .format(key, value) for key, value in zip(key.flatten(), con_m_train.flatten())])).reshape(2,2)
labels_test = (np.asarray(["{0} = {1:.2f}" .format(key, value) for key, value in zip(key.flatten(), con_m_test.flatten())])).reshape(2,2)

sns.heatmap(con_m_train, linewidths=.5, xticklabels=['PREDICTED : NO', 'PREDICTED : YES'], yticklabels=['ACTUAL : NO', 'ACTUAL : YES'], annot = labels_train, fmt = '', ax=ax[0])
sns.heatmap(con_m_test, linewidths=.5, xticklabels=['PREDICTED : NO', 'PREDICTED : YES'], yticklabels=['ACTUAL : NO', 'ACTUAL : YES'], annot = labels_test, fmt = '', ax=ax[1])

ax[0].set_title('Train Set')
ax[1].set_title('Test Set')

plt.show()
the maximum value of tpr*(1-fpr) 0.7586193479766012 for threshold 0.808
the maximum value of tpr*(1-fpr) 0.425603832860771 for threshold 0.86

2.5.2 Applying GBDT on Set - 2

In [89]:
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.model_selection import GridSearchCV
print(X_train2.shape, y_train.shape)
(20100, 9691) (20100,)

HyperParameter Tunning

In [85]:
GBDT = GradientBoostingClassifier()

parameters = {'max_depth' : [5, 10, 20, 50, 100], 'n_estimators' : [10, 25, 50, 100, 250]}

classifier = GridSearchCV(GBDT, parameters, cv=3, scoring = 'roc_auc')
classifier.fit(X_train2, y_train)                                         
Out[85]:
GridSearchCV(cv=3, error_score='raise-deprecating',
       estimator=GradientBoostingClassifier(criterion='friedman_mse', init=None,
              learning_rate=0.1, loss='deviance', max_depth=3,
              max_features=None, max_leaf_nodes=None,
              min_impurity_decrease=0.0, min_impurity_split=None,
              min_samples_leaf=1, min_sampl...      subsample=1.0, tol=0.0001, validation_fraction=0.1,
              verbose=0, warm_start=False),
       fit_params=None, iid='warn', n_jobs=None,
       param_grid={'max_depth': [5, 10, 20, 50, 100], 'n_estimators': [10, 25, 50, 100, 250]},
       pre_dispatch='2*n_jobs', refit=True, return_train_score='warn',
       scoring='roc_auc', verbose=0)
In [86]:
import seaborn as sea
max_scores = pd.DataFrame(classifier.cv_results_).groupby(['param_n_estimators', 'param_max_depth']).max().unstack()[['mean_test_score', 'mean_train_score']]

fig, ax = plt.subplots(1,2, figsize=(20,6))

sea.heatmap(max_scores.mean_train_score, annot = True, fmt='.4g', ax=ax[0])
sea.heatmap(max_scores.mean_test_score, annot = True, fmt='.4g', ax=ax[1])

ax[0].set_title('Train Set')
ax[1].set_title('CV Set')

plt.show()

Training Using Best Hyperparameters

In [127]:
# https://scikit-learn.org/stable/modules/generated/sklearn.metrics.roc_curve.html#sklearn.metrics.roc_curve
from sklearn.metrics import roc_curve, auc
from sklearn.ensemble import GradientBoostingClassifier

classifier = GradientBoostingClassifier(max_depth = 10 , min_samples_split = 250)

classifier.fit(X_train2, y_train)
#classifier_viz.fit(X_train1, y_train)

y_train_pred = batch_predict(classifier,X_train2)
y_test_pred = batch_predict(classifier,X_test2)
#y_test_pred = batch_predict(classifier_viz,X_test1)

train_fpr, train_tpr, tr_thresholds = roc_curve(y_train, y_train_pred)
test_fpr, test_tpr, te_thresholds = roc_curve(y_test, y_test_pred)

plt.plot(train_fpr, train_tpr, label="Train AUC ="+str(auc(train_fpr, train_tpr)))
plt.plot(test_fpr, test_tpr, label="Test AUC ="+str(auc(test_fpr, test_tpr)))
plt.legend()
plt.xlabel("True Positive Rate(TPR)")
plt.ylabel("False Positive Rate(FPR)")
plt.title("AUC")
plt.grid()
plt.show()
                                        

Train and Test Confusion Matrix

In [128]:
#https://www.quantinsti.com/blog/creating-heatmap-using-python-seaborn
import seaborn as sns
sns.set()

con_m_train = confusion_matrix(y_train, predict(y_train_pred, tr_thresholds, train_fpr, train_tpr))
con_m_test = confusion_matrix(y_test, predict(y_test_pred, te_thresholds, test_fpr, test_tpr))

key = (np.asarray([['TN','FP'], ['FN', 'TP']]))
fig, ax = plt.subplots(1,2, figsize=(12,5)) 

labels_train = (np.asarray(["{0} = {1:.2f}" .format(key, value) for key, value in zip(key.flatten(), con_m_train.flatten())])).reshape(2,2)
labels_test = (np.asarray(["{0} = {1:.2f}" .format(key, value) for key, value in zip(key.flatten(), con_m_test.flatten())])).reshape(2,2)

sns.heatmap(con_m_train, linewidths=.5, xticklabels=['PREDICTED : NO', 'PREDICTED : YES'], yticklabels=['ACTUAL : NO', 'ACTUAL : YES'], annot = labels_train, fmt = '', ax=ax[0])
sns.heatmap(con_m_test, linewidths=.5, xticklabels=['PREDICTED : NO', 'PREDICTED : YES'], yticklabels=['ACTUAL : NO', 'ACTUAL : YES'], annot = labels_test, fmt = '', ax=ax[1])

ax[0].set_title('Train Set')
ax[1].set_title('Test Set')

plt.show()
the maximum value of tpr*(1-fpr) 0.7628652361823036 for threshold 0.841
the maximum value of tpr*(1-fpr) 0.4244734776064234 for threshold 0.85

2.5.3 Applying GBDT on Set - 3

In [90]:
print(X_train3.shape, y_train.shape)
(20100, 610) (20100,)

HyperParameter Tunning

In [87]:
GBDT = GradientBoostingClassifier()

parameters = {'max_depth' : [5, 10, 20, 50, 100], 'n_estimators' : [10, 25, 50, 100, 250]}

classifier = GridSearchCV(GBDT, parameters, cv=3, scoring = 'roc_auc')
classifier.fit(X_train3, y_train)                                         
Out[87]:
GridSearchCV(cv=3, error_score='raise-deprecating',
       estimator=GradientBoostingClassifier(criterion='friedman_mse', init=None,
              learning_rate=0.1, loss='deviance', max_depth=3,
              max_features=None, max_leaf_nodes=None,
              min_impurity_decrease=0.0, min_impurity_split=None,
              min_samples_leaf=1, min_sampl...      subsample=1.0, tol=0.0001, validation_fraction=0.1,
              verbose=0, warm_start=False),
       fit_params=None, iid='warn', n_jobs=None,
       param_grid={'max_depth': [5, 10, 20, 50, 100], 'n_estimators': [10, 25, 50, 100, 250]},
       pre_dispatch='2*n_jobs', refit=True, return_train_score='warn',
       scoring='roc_auc', verbose=0)
In [89]:
import seaborn as sea
max_scores = pd.DataFrame(classifier.cv_results_).groupby(['param_n_estimators', 'param_max_depth']).max().unstack()[['mean_test_score', 'mean_train_score']]

fig, ax = plt.subplots(1,2, figsize=(20,6))

sea.heatmap(max_scores.mean_train_score, annot = True, fmt='.4g', ax=ax[0])
sea.heatmap(max_scores.mean_test_score, annot = True, fmt='.4g', ax=ax[1])

ax[0].set_title('Train Set')
ax[1].set_title('CV Set')

plt.show()

Training Using Best Hyperparameter

In [131]:
# https://scikit-learn.org/stable/modules/generated/sklearn.metrics.roc_curve.html#sklearn.metrics.roc_curve
from sklearn.metrics import roc_curve, auc
from sklearn.ensemble import GradientBoostingClassifier

classifier = GradientBoostingClassifier(max_depth = 5 , min_samples_split = 50)

classifier.fit(X_train3, y_train)
#classifier_viz.fit(X_train1, y_train)

y_train_pred = batch_predict(classifier,X_train3)
y_test_pred = batch_predict(classifier,X_test3)
#y_test_pred = batch_predict(classifier_viz,X_test1)

train_fpr, train_tpr, tr_thresholds = roc_curve(y_train, y_train_pred)
test_fpr, test_tpr, te_thresholds = roc_curve(y_test, y_test_pred)

plt.plot(train_fpr, train_tpr, label="Train AUC ="+str(auc(train_fpr, train_tpr)))
plt.plot(test_fpr, test_tpr, label="Test AUC ="+str(auc(test_fpr, test_tpr)))
plt.legend()
plt.xlabel("True Positive Rate(TPR)")
plt.ylabel("False Positive Rate(FPR)")
plt.title("AUC")
plt.grid()
plt.show()
                                        

Train and Test Confusion Matrix

In [132]:
#https://www.quantinsti.com/blog/creating-heatmap-using-python-seaborn
import seaborn as sns
sns.set()

con_m_train = confusion_matrix(y_train, predict(y_train_pred, tr_thresholds, train_fpr, train_tpr))
con_m_test = confusion_matrix(y_test, predict(y_test_pred, te_thresholds, test_fpr, test_tpr))

key = (np.asarray([['TN','FP'], ['FN', 'TP']]))
fig, ax = plt.subplots(1,2, figsize=(12,5)) 

labels_train = (np.asarray(["{0} = {1:.2f}" .format(key, value) for key, value in zip(key.flatten(), con_m_train.flatten())])).reshape(2,2)
labels_test = (np.asarray(["{0} = {1:.2f}" .format(key, value) for key, value in zip(key.flatten(), con_m_test.flatten())])).reshape(2,2)

sns.heatmap(con_m_train, linewidths=.5, xticklabels=['PREDICTED : NO', 'PREDICTED : YES'], yticklabels=['ACTUAL : NO', 'ACTUAL : YES'], annot = labels_train, fmt = '', ax=ax[0])
sns.heatmap(con_m_test, linewidths=.5, xticklabels=['PREDICTED : NO', 'PREDICTED : YES'], yticklabels=['ACTUAL : NO', 'ACTUAL : YES'], annot = labels_test, fmt = '', ax=ax[1])

ax[0].set_title('Train Set')
ax[1].set_title('Test Set')

plt.show()
the maximum value of tpr*(1-fpr) 0.6466596697976098 for threshold 0.85
the maximum value of tpr*(1-fpr) 0.4226376847460082 for threshold 0.863

2.5.4 Applying GBDT on Set - 4

In [91]:
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.model_selection import GridSearchCV
print(X_train4.shape, y_train.shape)
(20100, 610) (20100,)

HyperParameter Tunning

In [90]:
GBDT = GradientBoostingClassifier()

parameters = {'max_depth' : [5, 10, 20, 50, 100], 'n_estimators' : [10, 25, 50, 100, 250]}

classifier = GridSearchCV(GBDT, parameters, cv=3, scoring = 'roc_auc')
classifier.fit(X_train4, y_train)                                         
Out[90]:
GridSearchCV(cv=3, error_score='raise-deprecating',
       estimator=GradientBoostingClassifier(criterion='friedman_mse', init=None,
              learning_rate=0.1, loss='deviance', max_depth=3,
              max_features=None, max_leaf_nodes=None,
              min_impurity_decrease=0.0, min_impurity_split=None,
              min_samples_leaf=1, min_sampl...      subsample=1.0, tol=0.0001, validation_fraction=0.1,
              verbose=0, warm_start=False),
       fit_params=None, iid='warn', n_jobs=None,
       param_grid={'max_depth': [5, 10, 20, 50, 100], 'n_estimators': [10, 25, 50, 100, 250]},
       pre_dispatch='2*n_jobs', refit=True, return_train_score='warn',
       scoring='roc_auc', verbose=0)
In [91]:
max_scores = pd.DataFrame(classifier.cv_results_).groupby(['param_n_estimators', 'param_max_depth']).max().unstack()[['mean_test_score', 'mean_train_score']]

fig, ax = plt.subplots(1,2, figsize=(20,6))

sea.heatmap(max_scores.mean_train_score, annot = True, fmt='.4g', ax=ax[0])
sea.heatmap(max_scores.mean_test_score, annot = True, fmt='.4g', ax=ax[1])

ax[0].set_title('Train Set')
ax[1].set_title('CV Set')

plt.show()

Training Using Best Hyperparameter

In [135]:
# https://scikit-learn.org/stable/modules/generated/sklearn.metrics.roc_curve.html#sklearn.metrics.roc_curve
from sklearn.metrics import roc_curve, auc
from sklearn.ensemble import GradientBoostingClassifier

classifier = GradientBoostingClassifier(max_depth = 5 , min_samples_split = 100)

classifier.fit(X_train4, y_train)
#classifier_viz.fit(X_train1, y_train)

y_train_pred = batch_predict(classifier,X_train4)
y_test_pred = batch_predict(classifier,X_test4)
#y_test_pred = batch_predict(classifier_viz,X_test1)

train_fpr, train_tpr, tr_thresholds = roc_curve(y_train, y_train_pred)
test_fpr, test_tpr, te_thresholds = roc_curve(y_test, y_test_pred)

plt.plot(train_fpr, train_tpr, label="Train AUC ="+str(auc(train_fpr, train_tpr)))
plt.plot(test_fpr, test_tpr, label="Test AUC ="+str(auc(test_fpr, test_tpr)))
plt.legend()
plt.xlabel("True Positive Rate(TPR)")
plt.ylabel("False Positive Rate(FPR)")
plt.title("AUC")
plt.grid()
plt.show()
                                        

Train and Test Confusion Matrix

In [137]:
#https://www.quantinsti.com/blog/creating-heatmap-using-python-seaborn
import seaborn as sns
sns.set()

con_m_train = confusion_matrix(y_train, predict(y_train_pred, tr_thresholds, train_fpr, train_tpr))
con_m_test = confusion_matrix(y_test, predict(y_test_pred, te_thresholds, test_fpr, test_tpr))

key = (np.asarray([['TN','FP'], ['FN', 'TP']]))
fig, ax = plt.subplots(1,2, figsize=(12,5)) 

labels_train = (np.asarray(["{0} = {1:.2f}" .format(key, value) for key, value in zip(key.flatten(), con_m_train.flatten())])).reshape(2,2)
labels_test = (np.asarray(["{0} = {1:.2f}" .format(key, value) for key, value in zip(key.flatten(), con_m_test.flatten())])).reshape(2,2)

sns.heatmap(con_m_train, linewidths=.5, xticklabels=['PREDICTED : NO', 'PREDICTED : YES'], yticklabels=['ACTUAL : NO', 'ACTUAL : YES'], annot = labels_train, fmt = '', ax=ax[0])
sns.heatmap(con_m_test, linewidths=.5, xticklabels=['PREDICTED : NO', 'PREDICTED : YES'], yticklabels=['ACTUAL : NO', 'ACTUAL : YES'], annot = labels_test, fmt = '', ax=ax[1])

ax[0].set_title('Train Set')
ax[1].set_title('Test Set')

plt.show()
the maximum value of tpr*(1-fpr) 0.6345110308050191 for threshold 0.836
the maximum value of tpr*(1-fpr) 0.42250657276894876 for threshold 0.86

3. Conclusion

In [120]:
# Please compare all your models using Prettytable library
# http://zetcode.com/python/prettytable/
from prettytable import PrettyTable
TB = PrettyTable()

TB.field_names = ["Rand_Forest - MODEL", "HyperparameterS", "Train_AUC", "Test_Auc"]
TB.title = "Decision Tree"
TB.add_row(["BOW-ENC-RF", "Depth:10  | Samp_Split:250", 0.67,0.63])
TB.add_row(["TFIDF-ENC-RF", "Depth:10  | Samp_Split:250", 0.66, 0.62])
TB.add_row(["AvgW2V-ENC-RF", "Depth:5   | Samp_Split:500", 0.64, 0.60])
TB.add_row(["Tf-Idf-ENC-RF", "Depth:5   | Samp_Split:250", 0.65, 0.61])
print(TB)

TB1 = PrettyTable()

TB1.field_names = ["GBDT - MODEL", "HyperparameterS", "Train_AUC", "Test_Auc"]
TB1.title = "Gradient Boosting Decision Tree"
TB1.add_row(["BOW-ENC-GBDT", "Depth:5   | Samp_Split:250", 0.94,0.71])
TB1.add_row(["TFIDF-ENC-GBDT", "Depth:10  | Samp_Split:250", 0.92, 0.68])
TB1.add_row(["AvgW2V-ENC-GBDT", "Depth:5   | Samp_Split:50", 0.88, 0.67])
TB1.add_row(["Tf-Idf-ENC-GBDT", "Depth:5   | Samp_Split:100", 0.88, 0.69])
print(TB1)
+---------------------+----------------------------+-----------+----------+
| Rand_Forest - MODEL |      HyperparameterS       | Train_AUC | Test_Auc |
+---------------------+----------------------------+-----------+----------+
|      BOW-ENC-RF     | Depth:10  | Samp_Split:250 |    0.67   |   0.63   |
|     TFIDF-ENC-RF    | Depth:10  | Samp_Split:250 |    0.66   |   0.62   |
|    AvgW2V-ENC-RF    | Depth:5   | Samp_Split:500 |    0.64   |   0.6    |
|    Tf-Idf-ENC-RF    | Depth:5   | Samp_Split:250 |    0.65   |   0.61   |
+---------------------+----------------------------+-----------+----------+
+-----------------+----------------------------+-----------+----------+
|   GBDT - MODEL  |      HyperparameterS       | Train_AUC | Test_Auc |
+-----------------+----------------------------+-----------+----------+
|   BOW-ENC-GBDT  | Depth:5   | Samp_Split:250 |    0.94   |   0.71   |
|  TFIDF-ENC-GBDT | Depth:10  | Samp_Split:250 |    0.92   |   0.68   |
| AvgW2V-ENC-GBDT | Depth:5   | Samp_Split:50  |    0.88   |   0.67   |
| Tf-Idf-ENC-GBDT | Depth:5   | Samp_Split:100 |    0.88   |   0.69   |
+-----------------+----------------------------+-----------+----------+

Observations: </br>

1. From Above we can say that GBDT performs better compare to Random Forest models. Among them BOW encoded models will perform better, Since BOW is High dimensional compared to others. 2. As GridSearch for GBDT is Taking So much time i have taken 30K points as suggested. For those points the Random Forest and Gradient Boosting Decision Tree both performed well on BOW encoded features.